Contents
1. Transcript
1.1. Session One (May 20, 2013)
- FIDLER
- Let’s begin with your early life and background. How did your parents
arrive in New York, and what was their background in Europe?
- KLEINROCK
- So my mother came over first. She was four years old. She came over in
1911. Prior to that, her mother, who was widowed, took her eldest son,
came over to New York to basically create a small livelihood, generate
enough money to bring them over. They were extremely poor, living in a
city in Poland called Rzeszow. My grandmother worked for over a year,
finally was able to call the children over.The eldest daughter was in charge of those that were left behind in
Poland. The eldest daughter was my Aunt Minnie, and my mother became so
attached to Minnie, she was convinced that Minnie was her mother and had
a difficult adjustment to her true mother when she met her back here in
America. In fact, when they started to come over, they got to the port
in Europe to come over on the ship, and they got turned back because my
mother had some kind of a problem. She had a little hole in her head,
actually. It was some kind of disease. So they all had to go back.My grandmother had to generate some more funds. Finally brought them
over. So they settled in New York, and my mother grew up with her other
five siblings, very poor, extremely poor. I’m not sure exactly where
they were living. Part of the time, I think, was in Lower East Side, but
it was in Manhattan, in Jewish neighborhoods. My grandmother was a
midwife and she was—that’s what I remember about her. But the children all went to high school but, for example, my mother did
not go to get an academic high school degree. She went to get a
commercial degree and became a secretary. By the way, she was also a
beautiful woman, says I, and she actually did win a number of beauty
contests, and she beat out, in one of the local beauty contests, the
woman eventually became Miss America. So I’m very proud of my
mother. Now, my father, his story is a little less fortunate. He was born in
1905, and when the war broke out in 1914, his father was taken away into
the army. My father relates a story that one day his father was brought
to him in a wagon with other people, and his father said, “Son, I have
to go to war and I may never see you again.” And, in fact, my father
never did see him again, and it was traumatic for him.He was born in a city called Rudniki in Poland. It’s now the western
Ukraine. He was now living in a larger city called Podhajce, where he
was sent to live with his grandmother where he went to elementary
school. But when the war broke out, everybody started running away from
the Russians. He was living with his grandmother and his uncle. It was
just too difficult. They turned back to go back to Podhajce, and the
Russians were very good to them. But as the war worked on, by the time
1916, there was extreme poverty and almost famine and disease, and his
grandmother passed away. And then his uncle, who he was living with, was
taken off to war. His uncle managed to place him—I’m probably going into
too much detail, but—
- FIDLER
- Not at all.
- KLEINROCK
- His uncle managed to place him with a neighbor who promised to take care
of my father, and within a couple of months, threw him out because he
got into a fight with the other kids there. So now my dad was a waif in war-torn portion of Poland, which is being
passed back and forth between the Kaiser’s troops and the Russian
troops, and he lived with a gang of other kids, watering whatever horses
from whatever army came by and feeding them. In the summers, he was sort
of okay. In the winters, they were freezing. Soldiers were good to them,
but he was just a wild kid on his own, no education, Jewish kid, of
course, but the whole area was Jewish. He managed to get by. At the end of the war, his uncle came back, full of shrapnel, in 1919,
and within two years his uncle decided to go to New York, where my
father’s aunt, the uncle’s sister, had already settled. So he came here
in 1921 when he was sixteen, and he was a tiny tot of a guy because he
was seriously malnourished, and on the boat over, everybody was seasick,
but not him. There was a lot of food left over, so he filled himself up
pretty well. [laughs] He was thrilled. So he came here and he started living with his aunt. He started working
as a grocery clerk and attended night high school—well, elementary
school; he had no education—and very quickly raced through and got into
high school. But the effort was too much, working long hours, very
heavy-duty work, you know, carrying heavy boxes, long hours.And by this time, he was also thrown out of his aunt’s house. He knocked
out one of the sons. I’m smiling because I guess I’m proud of that, but
I shouldn’t be. [laughs] So he was now living on his own, and he fell in
with a bunch of guys who he spent time with and he palled around with
them. That was the phrase he used, “palled around” with them. And you
could see how much he needed companionship, having grown up in this
environment where he had no stability. So he dropped out of school and started hanging around with them, and it
was shortly thereafter that he met my mother. She was a secretary at the
time. And she asked him what he did, and besides being a grocery clerk,
he said, “I go to Columbia,” and that was true. He was going to Columbia
to go to the gym to work out. [laughs] But they hit it off and had a
long courtship. But, you know, Dad was working his way up as a grocery
clerk, but it was very difficult. They were very poor. So that’s
probably enough of that early history.
- FIDLER
- Was it a particularly religious or political family that you grew up in?
- KLEINROCK
- Neither, neither. They were very Jewish, very ethnic, and the language
from my grandmother was Yiddish. My father spoke perfect Yiddish. But,
no, it was not a religious family at all. It was not a political family.
They were struggling to get food on the table, make ends meet, and try
to put together a small life. And what my dad did is he—probably around the time—my parents got
married in 1930. My sister was born in 1932. I was born in 1934. It was
around that period when my dad joined with one of the fellow clerks, and
they decided to get enough money to buy a grocery store on 163rd and
Broadway in Washington Heights. And they struggled, and he grew that
store and made a success of it. He was extremely ethical, honest. He had
a reputation for being straight, and that helped him a great deal.He was able to borrow a little bit of money from his accountant, which
there’s a long story there as to how his partner said to him, “Listen.
We can’t continue as partners. Either I buy you out or you buy me out.”
And he knew my father had no money, so he figured he’d be able to buy my
father out. My father went to his accountant and borrowed, I think, about $1,000.
And when the time came, the partner said, “Can you buy me out?”My dad said, “Yes!” And he bought him out, and so he had his own store,
and he worked in different stores and did very nicely. And around that
time, I was born in 1934 in the midst of the Depression, of course.
- FIDLER
- Can you tell me about the neighborhood that you were born into and grew
up in?
- KLEINROCK
- So I was born in a hospital in Harlem, but we lived in Washington
Heights. That’s not far from the George Washington Bridge. It was a very
poor neighborhood. We were on the wrong side of the tracks, if you will.
Up in that area, if you were living west of Broadway, it was the good
neighborhood, and east of Broadway was the poor neighborhood. We went
from apartment to apartment, and the one I remember most is we lived on
175th and Amsterdam Avenue, and it was neighborhood where myself and one
other kid were the only two Jewish kids in a dominantly Catholic
neighborhood. In fact, there was a Catholic high school across the
street from where we lived. It was very difficult because there was a
lot of discrimination, and I was constantly running away from gangs.
They would sic their German Shepherds on me. But I grew up as a kid, you
know, you’re in the streets there. You grow up and you learn your way
around, and I had to grow up tough, take care of myself, and learn to
work the streets, you know, be careful where you go, how you get
cornered, etc., but this is all pre-school.I very much enjoyed my youth. It was very poor. I didn’t know we were
poor, if you will. My mother was home all the time. She was an extremely
warm, loving woman, thought the world of me, and that turned out to be
very important.But it was difficult. My dad was working. He was hardly ever home, and he
was very strict, and he came from the old school. Remember he grew up in
that tough arena, no parents, a bunch of wild kids, and the whole social
milieu of that part of the world was that children obeyed their parents,
and as soon as they could, they would have to contribute economically
and physically to the support of the family. And so my dad demanded a
lot of obedience.My mother was a softer one. But my dad was hardly ever home. He’d work
from seven a.m. to eight p.m. I’d be asleep often when he came home.
Except on Sundays, he worked a half a day. And yet I admired him
greatly. He was a strong guy. He had a large chest, terrifically strong
arms, and I remember he always had a pack of Chesterfield cigarettes in
the pocket of his T-shirt.But it was a difficult neighborhood. We were poor. My sister started
going to elementary school in the bad part of the area, so they decided
that when I would become elementary-school age that they’d move to a
slightly better area, because by now my dad was doing better in the
grocery store. So we moved just west of Broadway, one block from the elementary school,
PS 173, which I attended. So I did attend that school, and that was a
nice mix of smart kids, really smart kids in, if you will, gifted
classes. So I was in many ways separated from the harsh environment of
the tough, unruly, disobedient, and nasty kids. A lot of Jewish kids, a
lot of non-Jewish Christian children, all smart, and it was a terrific
school. I really loved school.Until I went to school, I’m told I was a really wild kid. My mother
couldn’t take me into a store because I’d break things, you know,
bumbling around, touching things. They would tell her to take her
business elsewhere. But somehow when I got into elementary school—I
remember this, and it’s strange, where I somehow changed and I became
milder, more obedient, because I was now in a structured environment
where there were clear expectations put on me by the environment at the
school, you know, no nonsense. And I liked it because it was structured,
there were things to do, there were challenges, it was clear what was
expected of me and what I had to do, so I somehow warmed up to that very
well.And, you know, I was with a bunch of guys, but I still was, quote,
“different” in my own mind because I was poorer than most of the kids.
They had been living in the good area all the time, they had nice
clothes, and they had an environment where kids and other adults would
come to their apartments. It was not the case in my family. There wasn’t
a flow of people. I never brought my friends to my house.
- FIDLER
- And this is during PS 173?
- KLEINROCK
- During PS 173.
- FIDLER
- Because another question I was going to ask was how did you understand
your socioeconomic background as I child, and I assume that changed
somewhat over the years. But by the time you were at PS 173, you did
understand something about having a particular background, a particular
status.
- KLEINROCK
- Yes. I always felt poorer than the other kids. I saw what they had. I
saw the things they did. I saw their parents were more engaged. I
remember after school some of the fathers would come out and they would
toss a football around or play punch ball with their kids. I had none of
that. And occasionally I’d visit their homes, and it was different. It
was neater; it was organized; it was sane. Ours was a little more hectic
and unruly in a way. But I did feel different because most of the boys were very much into
baseball, professional baseball. They would know all the players and the
stats and the games and the scores. And not only didn’t I know, I really
didn’t care. I wanted to play softball, you know, go out and play
football, not talk about it. So we had little clubs. I was in a group.
We called ourselves the Avengers. We had a baseball team. It was
actually softball. Had a great time. But they spent most of their time
talking about those stats, exchanging baseball cards, and somehow that’s
not where my interest was. I was more interested in reading books about
astronomy. I’d go to the library, sit up all night. I was interested in
science very early on. In fact, the things that interested me then were
gadgets, puzzles, games, and comic books. I was really into comic books
a lot. I just knew I had a different level of interest and a different
socioeconomic background, as you say. And then I was eleven, I was in fifth grade, my father got extremely
ill. In fact, we didn’t have a separate bedroom for the kids when we
were living on Broadway. My parents slept in the bedroom. My sister and
I slept in the living room. My father got extremely ill, and he was
coughing all this one night. Next day, I went to school and when I came
home, I saw them taking my father out into an ambulance and was taken
away to the hospital, and he had a serious, serious case of asthma and
bronchitis, and it really took him down. I was eleven years old. He had a grocery store by that time, as I said.
He owned one, and he had done very well. This is now at the end of the
war in 1945. So he’d done well in the war, but suddenly he was earning
nothing. We had to live on whatever savings he had. He couldn’t work. He
had to sell the store. My mother was not working. Shortly thereafter, we moved back on the other side of Broadway to St.
Nicholas Avenue. We lived there until I left home, spent many, many
years there. It was in an apartment with one bedroom, which my sister
had, and I slept in the living room on a fold-up cot. That meant every
morning I’d have to fold it up again.My dad was at home ill. In the first apartment, there was a bedroom that
my sister and I shared. I remember that. I was in the cot in the second
apartment. And their bedroom was the converted living room with no door,
so he demanded absolute quiet. So imagine young kids having to be very
quiet. It was really oppressive, but we had to obey.There was no money coming in. They were getting worried about money. It
was leaking away. And yet I was taking violin lessons, had to practice
an hour a day. I was a lousy violinist, by the way.And four o’clock every afternoon after school, I’d go to Hebrew school.
They sent me to a religious Hebrew school, even though they were not
religious. So I had from three to four in the afternoon to pal around
with my friends, and then from then on it was all Hebrew school, violin
lesson, homework, off to school again next day. And because they needed
some money, I started working as a stock boy in a butcher store and then
in a children’s clothing store early on. All the money I made, of
course, I brought home and gave to my parents.
- FIDLER
- And about what age did you start working?
- KLEINROCK
- I can’t say exactly. It was probably twelve, thirteen. It was after my
dad got sick and before my mother went to work. She went to work a few
years later, which was a big blow to us, my sister and I. You know, she
had been home all the time. She went to work in Barton’s Candy Store as
a clerk and very quickly rose to the level of manager. She was a
wonderful worker. And when we’d go to visit the store, she’d give us
samples. [laughs] And we didn’t mind that one bit. So the environment, it was very hectic. I didn’t even have a desk to do
my homework on, and I was working a lot at that point now. By this time,
I was in junior high school. From elementary school—I loved elementary
school. It was terrific, learned a lot. Junior high school was a kind of
place where all the elementary schools fed their students, so many
elementary schools per junior high. This is back in the bad part of the
neighborhood, and it was a hellhole.
- FIDLER
- And this is Humboldt?
- KLEINROCK
- Humboldt Junior High School 115. It was 177th between Audubon and St.
Nicholas.
- FIDLER
- When you went there, just to back up slightly before we get into
Humboldt, your group of friends, did they change when you got to junior
high?
- KLEINROCK
- No. We went as a group, so I was with many of the same kids but many
others as well, and you had some really wild kids there, you know, gangs
all over the place. When I went back there just a year or two ago, that
whole place is boarded up with fences and guards. It’s still a bad
place. It was hard, but the education I got there was okay because we
did have these advanced classes, nevertheless. But it’s just the
lunchtimes and the physical ed, and the breaks, that were hard. You had
to watch out what you were doing.
- FIDLER
- And did you have any choice at all in terms of your eventual move to the
Bronx High School of Science and Mathematics, or was that an automatic—
- KLEINROCK
- Oh, no. For Bronx Science, I was destined to go to the public high
school. Well, they’re all public high schools, but the one I was
destined to go was George Washington High School, where my sister went.
But there were these specialized high schools in New York at the time,
for which you took exams. By then, I was very much interested in
electronics, and so I and a number of my other classmates took the exam
to get into Bronx Science, an exam you had to take, and I was amazed
that I got in. I didn’t expect to get in. It was two subway trips away
to get there instead of a bus ride up to George Washington, and I was
thrilled to go there. A lot of other kids in my class who I thought were
smarter than me didn’t get in, and I was surprised. But it was a great
move.But by then, I was very much into electronics. When I was much younger,
when I was just entering elementary school, as I told you, I was very
much interested in comic books, and I saw in the middle of a Superman
comic how to build a crystal radio. In fact, I was about six or seven
years old at the time, and there was a description in the centerfold
which talked about how to build this crystal radio. What fascinated me
were two things: one, I could build it out of parts that I knew I
wouldn’t have to pay a penny for because I could find them around the
house; and secondly, it claimed that I’d be able to hear music out of
this crystal radio without any batteries, any electricity. It was sort
of all free.So I decided to do that. I loved gadgets and puzzles, and this looked
like a challenge. So what did I need? I needed an empty toilet-paper
roll, which I could easily get at home. I needed some wire, which I
could find in the street, to form an inductor, wind it around the
toilet-paper roll. I needed a crystal. Well, they explained you can make
a crystal out of your father’s old razorblade and a piece of pencil
lead, instead of buying one. That formed what’s called the diode, which
you needed for this. And then I needed an earphone. Now, I didn’t have
an earphone, but I knew in the candy store across the street there was a
telephone booth, and I knew you could unscrew the earphone of these
telephones, headsets, handsets, and take it away. So I stole that
earphone. Still free. Then I needed one more part, which I knew I didn’t
have around the street or the house, and that was something called a
variable capacitor. So I had my mother take me down to the place in New
York where they sold all the electronics parts, down to Canal Street. I
walked up to the first electronics store with my mother, and I banged my
fist on the table and I told the proprietor, “I need a variable
capacitor.” And he said, “What size?” And it totally blew my cover. I explained to
him exactly why I needed it, and I was a little embarrassed that I
didn’t know what size. He knew exactly what I needed. He sold it to us
for about a nickel. It did cost a nickel. Brought it home, wired it up,
and, lo and behold, I heard music as I tuned that capacitor. I could
change stations. And I was amazed. I mean, this was literally magic. It
still is magic, by the way. Anytime you have force at a distance, it
really is magic. Electricity, electromagnetism, radio, it’s a wonderful
gift that nature has given us. And so I spent the rest of my life trying
to figure out how that thing worked. I understand it, but there’s still
something magical about it. So what I did at that early age is I began to collect old broken radios
and cannibalize them. My cousin had a good friend who was in charge of a
television store, and there had been a repair shop back there. They had
a lot of old broken radios. I got them all. I filled up a closet full of
these old vacuum-tube systems, and I unsoldered them and I started
making new radios. So I was very much into electronics as a kid.
- FIDLER
- And was that pursuit a solitary one, or were there different people from
different parts of your life that would encourage or participate?
- KLEINROCK
- That’s a great question. Pretty much most of my life as a youth, I’d
been a loner. So when I started doing this radio work, it was all on my
own. Now, it would have been far more natural to find some other kids
who were interested in this and form a club, share ideas, share parts,
experiences. Didn’t do it that way. I used to build model airplanes a
lot, too, on my own, not with other kids, which is interesting.Again, this notion that you referred to earlier, I felt isolated as a
kid. I didn’t feel part of a group. When I built the model airplanes,
for example, I could look out the back window of my apartment into a
courtyard, and on Sundays these kids would come by with their model
airplanes, but they could afford gasoline engines, and they’d test them
out there and they’d spin those propellers and go [demonstrates]. I
couldn’t possibly afford an engine like that, so mine were all
rubber-band-based. Similarly with the radios; it was on my own. I think
I missed a lot. But as a result, when I went to Bronx Science, I didn’t realize I was
destined to be an engineer. In fact, I was resisting the idea because I
like to keep my options open. And I was a little reluctant to go to
Bronx Science, figuring I would now be channeled to become a scientist,
instead of possibly becoming something else which I couldn’t even
imagine, but I decided to go because it was the best school in the
country at the time. So I went, and one of the first classes I took, among others, was a
social studies class, and I was thrilled because now I’m going to get
more than just science, except when I walked into the class the first
day, the teacher said, “We’re going to study social studies using the
scientific method,” and that really worried me. But in fact, it was a
regular social studies class full of the real stuff.
- FIDLER
- And why did that worry you? Can you expand on that at all?
- KLEINROCK
- As I say, I like to keep my options open. I wasn’t committed to be an
engineer, nor even a scientist. Maybe a doctor, maybe a lawyer, maybe a
pilot. I wanted to be a pilot, by the way. Maybe an explorer. Something
else not channeled. And yet little did I realize that by that time in my
life, it was already ordained that I would become an engineer. You know,
all my interests were going in that direction. I remember in Bronx Science one of my classmates and I were on an
elevated train platform, waiting for the train to come. He was a real
nerd from Bronx Science. And I looked down. We were probably about 100
feet high on this elevated platform, and I said, “I wonder how high we
are.” So he said, “Oh, just a minute.” He reached down into the gravel, picked
up a stone, dropped it, timed it, and told me exactly how high we
were.I said, “How’d you do that?” He explained a little physics principle, and
I found those kinds of practical uses of science just intriguing. So I
took physics as one of my first courses at Bronx Science, way before
chemistry, which is the opposite order, by the way. And physics, well, I
just loved physics. In many ways, you know, electrical engineering and
physics are coupled. They’re the same kind of basic studies. But you’re right, not only was I a loner, but I wasn’t even socially—I
won’t even use the word “advanced,” but socially adept. Give you an
example. In high school, I wore orthodontures, braces for my teeth, and
I remember I was probably a junior in high school, one of my friends
said to me when he saw these braces on my teeth, says, “How do you kiss
the girls?” And my answer was, “I’ve never kissed a girl.” [laughs] So I was really
backward and socially backward at the time. I lived in a kind of not
economically protected, but a kind of sheltered environment from—you
know, I was taking care of myself, managing in this world I found
myself, but not participating in it in a way. And by the way, just to jump ahead, when I look at the colleagues that I
later interacted with in my professional life, I find many that had
similar backgrounds. They were not the guys who traded baseball cards.
They weren’t into sports statistics. They weren’t those kind of
collectors. They were often on their own, doing their thing, doing it
well, and not making a shared effort out of it. And I actually got a lot
of gratification out of that when I realized later I was not unusual in
that sense.
- FIDLER
- I was going to ask how this outsider status affected your intellectual
development, but that seems to answer at least a bit of it.
- KLEINROCK
- Well, it also made it important to me to succeed based on my own wits,
and not asking other kids how they did their homework. I remember later
in college when we did lab reports and you had to write them up, and
they were a long and tortuous thing to do, I never took data or advice
or information from other kids. I wrote all my own lab reports—the other
kids were copying them and sharing them, and there were older ones they
used—only because that’s the way I felt I had to learn, and I just
didn’t like to use other people’s material. I needed to do it myself so
I’d learn.
- FIDLER
- Did you continue to work part-time through your time at Bronx High?
- KLEINROCK
- Oh, yes. I was working many, many hours. I took a job as an usher
because we needed the money. So when I’d get home from Bronx Science,
I’d immediately rush off to usher. I was an usher in two movie theaters
at the same time because they were owned by the same corporation, and
I’d work there typically from about four o’clock until seven or eight in
the evening during the weekday, and on the weekends, longer hours. So I
didn’t have much time for play. It was not a difficult job, but it was time-consuming, and I learned a
lot about life in that job. Again, there were a lot of gang members,
also, in my fellow usherette, and I befriended them and I learned about
how they operated, how they lived, and in some ways, they protected me
because I was sort of a buddy of theirs. One of the usherettes was a prostitute, and she ran tricks in the
balcony during the theater. And just to tell you what was going on
there, we’d have to get out of our city clothes and get into these ugly
ushers’ uniforms, and we just would strip down to our shorts in an open
area with the girls and the guys, and every so often this usher would
show me the fancy embroidery on her girdle. One day she and I had an
argument. She was older than I was, quite a bit older. And then she
looked at me. She said, “You know, I would have given it to you for
free.” And my response to myself was, “What is she talking about?” [laughs] I
was that naïve at that time. I was really not with it.
- FIDLER
- And this is during high school?
- KLEINROCK
- Yeah, during high school. But I worked the whole time in high school.
- FIDLER
- And between, for example, the advanced classes at Humboldt and your
classmates at Bronx Science, was there a shift, do you remember?
- KLEINROCK
- Considerably. When I got to Bronx Science, I was surrounded by all
smart, somewhat nerdy, aggressive, middle-class, good-background kids
who were prepared. They had the kinds of backgrounds I was not
privileged to have. You know, they had parents who were helping them
with their science work and their projects, made sure they did their
homework. I felt like an outsider there, and they were richer than I was
and they were clubby, sometimes to exclusiveness. I was scheduled to graduate in January of 1952, but I knew that if I’m
going to go to college, entering college in the middle of the school
year is a bad idea. So I decided to accelerate and skip one grade, so I
would graduate in June of ’51 and be on schedule for proper school
entrance. That was a mistake, because not only was I not close to the
social groups, but even then I shifted my class, so I went from one
group to another group. I was neither fish nor fowl in either group now,
and so, again, I had one or two friends, but never in a close-knit
group. Anyway, as soon as school was over, I had to rush home and do my
work as an usher.So it was an isolated, independent, do-it-yourself kind of existence I
had there, and learn whatever you encounter, but not by the experience
or the benefit of how others had done it, which was unfortunate and made
it more difficult.
- FIDLER
- Can you tell me about your time in the Scouts and in particular your
experiences getting to and being an Eagle Scout?
- KLEINROCK
- It’s a wonderful time. When I was a kid—it was shortly after my father
got ill—I was very interested in joining the Sea Scouts. Now, to join
the Sea Scouts, you had to be four-foot-ten, and I was not. So every
Friday night, I would go to their meeting and asked to be measured, and
I became a bit of a nuisance. I remember one day when I went there, it was
snowing, so I had to wear galoshes, and when I got to make my
measurements, I decided to stand on my toes inside my galoshes, and I
passed the four-foot-ten mark. They let me in. I became a Sea Scout, and
I loved it because you could march around with wooden rifles in the
Armory. I really was interested—you know, this was just after World War
II, and it was a time of the military, and the whole patriotism and the
glory was a big deal for kids my age. I did know all the fighter
aircraft, all the planes all flying overhead.
- FIDLER
- What year was that, and about how many times did you get measured before
you—
- KLEINROCK
- I’d say five or six times. I’m calculating it was just before I was
twelve years old, because at twelve you could join the Boy Scouts, and I
think Sea Scouts allowed you in just as you were approaching twelve. So I got in and I loved it. Oh, I learned about knots. I had a sailor’s
uniform. It was a complicated outfit with all the buttons, the stripes,
and the sailor’s hat. And you learned about ships and knots, and we
actually got to march in the Saint Patrick’s Day parade up Fifth Avenue.
Now, that was a big thrill with the bands and all. It was a big kick.
But that Sea Scout group disbanded about a year after I joined, and I
was disappointed, because I just loved it. I decided to now get into the Boy Scouts. So I came in, as everybody
else, as a tenderfoot, and the reason I wanted to join these things, not
only for the military side, but I knew the Boy Scouts did camping.
Growing up in New York City, on the concrete sidewalks of New York City,
I loved nature. I wanted to be Tarzan as a kid, and if I couldn’t be
Tarzan, I wanted to be an American Indian. I wanted to get out there,
ride a horse, shoot a bow and arrow, climb a tree, and there was none of
that. But I knew the Boy Scouts did go out there. They’d go hiking,
they’d go camping, they’d make fires. So I joined the Boy Scouts, and I just loved it. I loved it for a lot of
reasons. First of all, again, just like school, it was structured. There
was a clear order of procedure. There was a ranking. And just like in
school, as you progressed, you had marks of achievement, and those were
important to me, clear milestones. You’ve achieved this, you’ve achieved
that from tenderfoot to second class to first class, and I worked my way
up the ranks, and I became a member of a patrol, then I became a patrol
leader, then I became the senior patrol leader. So I basically was in
charge of all the patrols, which is a wonderful experience in learning
about management and executive and leadership, which is what the Scouts
were all about. So I became what’s called a Star Scout, which means you’ve earned five
merit badges. I really enjoyed camping. All the other kids had sleeping
bags. When we were camping, I had a blanket, froze my behind off every
night.
- FIDLER
- Were you doing that, again, as an outsider, or were you finding a sense
of belonging by this point?
- KLEINROCK
- No, I felt a real sense of belonging with these kids. Every Friday
night, we’d meet, and I had great leaders. I looked up to them. I became
a leader. I got the respect. I got the chance to take a leadership role
and, if you will, a collaborative group experience, which is excellent,
and I just loved going out camping overnight and doing physically
challenging things and gaining skills. So when I became a Star Scout, my Scout master, whose name was Mr.
Skinner, I remember very well, he said to me, “Len, if you want, you can
become the first Eagle Scout in this troop. It’s not easy, but I have
faith that you can do it.” And I was very flattered that he said that to
me, so I decided, yeah, I’m going to try it, and that was a goal that
was basically just outside my grasp. You’d have to get twenty-one merit
badges. In those days, it was an arduous task. There were no Eagle
Scouts in my troop. They were like the glorified, honorific people in
the whole scouting tradition.I said, “Why not,” because I love Scouting. So I set out to do it. I
became a Life Scout, which is ten merit badges, and then to get the
other eleven merit badges—actually, it was before I became Life. I knew
it was very slow progress getting merit badges. Most of the merit badges
you had to get out in the wild on a camping trip. So I decided to go to
Scout camp one summer, and it was a two-week camp, and in that two
weeks, I got to thirteen merit badges, which was literally unheard of. I
was busy all the time, working like hell. So I got all but one of the
merit badges I needed for Eagle. The one I couldn’t get because it just
took too much time was Bird Study. That meant you had to go out every
morning. You had to see, understand, identify, know all about forty
different birds. By the way, they’ve dropped that requirement now from
the Eagle Scout. So I had to go back the next year just for that one
merit badge. I got it, became an Eagle Scout. That was a great milestone in my life, because it was an example of
something that was clearly not easy to get; it was a challenge. It was,
as I say, just outside my grasp, so I had to really reach to get it, and
I achieved it. And when I did that, it was different from getting 100 on
an exam. Other kids get 100, but I was the first to get that Eagle
Scout, and I realized that if I set my mind to it, I could achieve the
things I wanted to get, arduous and difficult though it may be. That
sort of set a—not a standard, but an approach, a kind of confidence,
that, yeah, if you want to do it, you can do it.So while I was in high school, I did become an Eagle Scout, and it was
very important to me. In fact, that Eagle Scout achievement means so
much to me, that since then in my professional career, I’ve gathered
many, many other honors, my bachelor’s degree, my master’s, my Ph.D.
degree, all kinds of honorary societies and best papers and fellowships
and fellow members. I’ve got a collage at home where I took all those
diplomas and I put them in a collage, and the one on top is the Eagle
Scout, and another one that’s peeking out is my Karate Black Belt, my
Nidan Black Belt. So it’s those kinds of nonscholastic achievements that
mean most to me, but the scholastic I did, but it’s those others that
represent a side of me that I see myself as, not as the professor, but
as the all-around person. In junior high and high, I studied Latin, and there’s a wonderful phrase
in Latin—I’d have to look up the Latin. I forget it, but what it means
is “A sound mind in a sound body,” and I took that to heart, and it’s
true. If you keep yourself physically fit and capable, it clears your
mind. They work together and they each benefit each other. So, end of
Eagle Scout story.
- FIDLER
- You’ve mentioned elsewhere that prior to your Eagle Scout status, you
were a good student but not a top student, and this helped you achieve
more scholastically, in fact.
- KLEINROCK
- Yes. I was a very good student in elementary school, junior high, high
school. I was never the top of the class. I was right up there, but I
didn’t see myself as a super achiever. When I got the Eagle Scout, it
made me think differently. I knew I could do a lot more, and later in
life I was able to become the best of class, if you will. I remember one rather interesting story in elementary school. I was
really good in math and science and not so good in English and
penmanship or conduct. I’ve seen my report cards since. It says “Needs
improvement in conduct and penmanship.” But the math and the science, I
would ace those things. So I remember one day the teacher in elementary school gave us a kind of
verbal arithmetic puzzle, and she would say, “Take this number, add such
and such, multiply it, divide, add,” but a long sequence. Then she went
around the room in order and asked, “What’s the answer?” She went up and
down the rows. She got to me and I said, “Naught.”She went around and around, and everybody else, she says, “Wrong, wrong,
wrong.” She got to me, she said, “Wrong.”She got to somebody. He said, “Zero.” She said, “Right.” I said, “Wait a minute. I said naught.” She didn’t quite understand the
word somehow. [laughs] Now, I don’t know why I said “naught,” but I was
pleased that I got it right. Whenever you score well, it just builds
your own confidence.
- FIDLER
- Let’s move to your time at the City College of New York. What was it
like being there in the 1950s?
- KLEINROCK
- Let me talk to you about how I got to City College. I was doing fine in
Bronx Science, made sure I graduated in June of ’51, but I knew that if
I’m going to go to college, I really wanted to go away to college and
live in a dormitory and have the full college experience. It was kind of
idealistic—see, most of my classmates at Bronx Science were planning to
go away to whatever, the Ivy Leagues or liberal arts colleges, even
Columbia, and some were going to City College. But I decided the only way I can go away to college is if I get a full
scholarship, because we just didn’t have the money. My dad was not doing
anything and my mom was struggling to earn a living. So I wrote away to
almost every Chamber of Commerce in the United States, and I asked them,
“Which of your schools do you have scholarships for,” and what kind of
pay were they giving. And I got a whole bunch of scholarships. I wrote
away and I applied. The trouble is none of them was sufficient.
Typically, they pay for room and board and tuition, but the sad fact is
there were still some expenses, the travel there, the travel back and
money to help the household. Couldn’t make it happen. So I decided I’ll go to City College, which was basically free and a
fantastic place. City College was one of the best colleges in the
country at the time, but those were the golden days when all the
immigrant kids, those who couldn’t afford to go to Columbia or NYU or
out of town, went there, and those immigrant kids or children of
immigrants, as always, were highly motivated, very, very studious,
disciplined, goal oriented, and they weren’t going to fool around. So it
was a great place to go. So I was all set to go. That summer of 1951, I took a job as a lifeguard—I was on the swimming
team at Bronx Science, by the way. I just loved that, sound mind and
sound body kind of thing, and I took a job as a lifeguard in New York
City. The pool that I was a lifeguard at was in the Lower East Side, by
the way, which was rather interesting because I broke up more fights
than I saved lives, and I saved fourteen lives. But it was, again, a
life experience. I remember the head of our lifeguard group, he was an experienced
lifeguard, and this particular pool was full of gangs. It was just full
of gangs. It’s 1951. So the first thing he did the day we got there is
he walked over to the leaders of the two top gangs, and he made peace
with them. He said, “Look, I’ll give you some rope, but you’ve got to
help me keep order and discipline here.” He made an alliance. Boy, was
that something to learn, really an experience. So anyway, I was a lifeguard there, and during that summer I was set to
go to City College that fall. Toward the end of that summer, my dad took
me to visit one of his cousins; that was one of the cousins in whose
home he lived when he came here to New York with his aunt. He’s not the
one that my father knocked out; it was another one. This guy was a
brilliant electrical engineer, and he had a very small firm called
Photobell Corporation, which was building industrial electronic devices.
My dad took me down there, and Abe Edelman was the owner, my father’s
cousin. He showed me around. It was a fascinating place. They were
building electronics from scratch. They would make the chassis, cut the
holes, put the vacuum tubes in, wire up the resistors and capacitors,
conductors, put in the power supplies. It was a fascinating place. And little did I know that my father had an ulterior motive; he wanted
me to work there instead of going to school. And Abe offered me a job.
When my dad and I came home, my dad basically said, “Len, you really
should work there.” And I said, “Dad, I’ve got to go to college.” He said, “Well, you know, we really need the money.” So he convinced me
to go to night session instead of day session, and work at Abe’s
place. Now, that was a huge blow to me, but I didn’t even realize it at the
time. I sort of took it in stride and said, “Well, I guess I’ll do
that.” I didn’t realize in some sense the danger I was being put in,
because most kids didn’t graduate evening session. They’d drop out or
fail out, and I feel especially bad about that because he had the same
experience. He couldn’t get through high school at night for the same
reason, the pressures of work and all the rest. But anyway, he sent me
there, and I have a deep resentment for that situation. My mother in
some sense didn’t object either. She didn’t push it. So I went to
evening session, and happy-go-lucky, I was ready to take on anything,
and so I did.So I went to work there, and it turned out to be an amazing experience. I
couldn’t go to college full-time, obviously, but I decided to push it as
hard as I could, so I went three-quarters time and every summer and
worked daytime, so I’d be up at six-thirty a.m., down to work, get to
school by close to six p.m., take my classes, finish typically at eleven
at night.And then, just because I was desperate for some break in that schedule, I
participated in what substituted for fraternities at City College; it
was called House Plan. They were clubs. They were brownstone buildings
next to City College. We’d go there and sort of hang out with guys and
gals, and there were guys’ clubs and gals’ clubs, and it would typically
close at midnight. So I joined this group called Dean House, a bunch of
Jewish and Italian guys. We had a great time, wild guys, you know; I was
young and vigorous and full of spit and polish at the time. Then I’d get
home at about twelve-thirty, one a.m., and get up early in the morning.
Now, I did my studying on the subway going to and from work. So what was this environment? I was, at night, with a very strange group
of people, a bunch of losers who couldn’t get into day session, who
would drop out, a bunch of poor kids struggling, some of whom could make
it and some of whom couldn’t, and a bunch of GIs, having come out of
World War II on the GI Bill, and these guys were really motivated to get
a degree. They knew what life was about. They knew they needed this
degree. And there were some kids like me who were too poor to go day
session. So it was mixed bag, and I felt sort of out of it, in a way,
because I wasn’t on a track to go to college like anybody else. Now, when I was in high school—I told you I studied the violin—I was in
the high school orchestra. It turned out one of the first mathematics
classes I took at CCNY at night—it was the first class—was taught by my
orchestra teacher from high school—you know often music and mathematics
go together—and I was so thrilled to see him. It was like having a
friend in an otherwise foreign environment. So that was a certain sense
of familiarity, encouragement. So those are my classmates. These are a
lot of hardworking kids, some of them smart, some of them not, but in
electrical engineering you didn’t find a lot of dumb kids. They’d be in
the softer classes, the kids who couldn’t handle it. There were a lot of gangs preying on evening-session classes. I still
have my calculus book from a class I took on the ground floor of Shepard
Hall at CCNY. It was a hot summer night class. The windows were open. A
gang came by with buckets of water and tossed water into the class, and
my pages of the book—it’s over here, actually, those pages are still
warped from where the water hit. But there weren’t any physical
encounters except going to City College at night. It was in Harlem, so
you had to be careful where, what time, what street you used. And all
during that period, and still to this day, I tend to carry my briefcase
in my left hand, so my right hand is available. So anyway, that was at school. Now, at work, what did I have? I had
electrical engineers, technicians, a really tough boss—Abe Edleman was a
tough guy—an inept manager. There were about eight to ten of us. He
never could make a strong living at this. He wouldn’t charge enough for
his product, but he was a brilliant engineer. I learned a great deal
from him. In fact, I remember when I first met him, he gave me a little
engineering problem. I won’t describe it to you, but he asked me how do
you calculate the resistance of this thing when you move this around.
And I got it. But what I had, they were engineers doing electronics, and
I was learning at an enormously fast pace. I learned what the real world
was about. The things you learned at electrical engineering school you
could apply here and vice versa; things you learned on the job you could
bring to the classroom. One of the first things I did when I was still a naïve student, having
just gotten out of high school, working at Photobell, I’d taken an
excellent class in descriptive geometry at Bronx Science. I really knew
how to draft and do descriptive geometry, which is really intersecting
solids and seeing what the intersection is, very well. So Abe, as one of
my first jobs, said, “Look. Take this phone plug, solder a resister
inside of it, put it together, and make a drawing of what you did.”
Simple enough. So I did that and I made a drawing, but I made a
draftsman drawing. I measured every dimension, put in perfect arrows. It
was a very nice eight-and-a-half-by-eleven picture of this thing.I gave it to Abe, and he nearly killed me. He said, “What are wasting
your time doing that? I just wanted a sketch. You wasted about an hour
that you took to draw that.” So I quickly learned about the practicality
of the real business world. The mix of that practical experience and the theoretical learning at
school, whereas at the same time, those professors themselves were in an
environment like me, they were working during the day and teaching at
night, and they brought to this class their practical experience. The
example I like to use here is I remember one of my teachers came in and
he held in his hand a funny little device. He says, “See this? This is a
transistor.” They’d just been invented. And he said, “They’re used as
amplifiers in place of vacuum tubes.” And he said, “However, they’re
better thermometers than they are amplifiers because they’re so
heat-sensitive. And what you do because they’re heat-sensitive is you
have to put this into the circuitry to combat that heat variation.” Now, I know that no day-session professor would ever have understood
that issue about the heat sensitivity or explained to the students how
to deal with it. So I had the advantage of practical experience, good
theoretical people teaching me, getting the practical experience at the
work, around GIs who set the tone for what you really are in school for,
not to play, because there was no play. It was a serious business, and I
basically set my course and I managed to graduate in five and a half
years, which is kind of an achievement, because the program for day
students was five years. In engineering, you took five years. It was 142
credits. I did it in five and a half because I was going every summer,
taking as heavy a load as I could. When I first got in, I joined Dean House, as I told you I was just out
of high school on the swimming team, and I decided, “Look, let’s make an
evening-session swimming team, and we’ll challenge the day-session
swimming team.” And so I got this ragtag group of guys from Dean House,
who were not really swimmers, and they all volunteered, “Sure, we can
swim!” And I was a good swimmer. So we put this team together. We never
practiced. We challenged the freshman team. Little did I know the
freshman team is enormously strong. So we had this meet, and I was a sprinter, so I swam the 50-yard sprint
and I think I came in second. One of our guys says, “I can swim the
220.” That was the long-distance race. So when we got to do it, he says,
“No, I can’t.” And nobody else would volunteer. So I said, “Well, I’d
better do it.” So I sprinted the 220. Well, you can’t sprint the 220. By
the time I got to 150 yards, I began to fade, and by the time I got to
200, I was almost going backwards. So we lost that meet. But the point is, I did put together a swim team. I had some of the
social aspects. I even became evening-session student president, and I
graduated first in my class, day and evening, of electrical engineering.
So, I managed to do well.
- FIDLER
- The house plan and the swim team, that would be the group of people
outside of classes that you would interact and associate with?
- KLEINROCK
- Well, the house plan and swim team were the same guys. We had a great,
great time, and my social life was basically the house-plan guys and,
again, it was a bunch of—a mixed group. There was, I think, one other
engineer. There were some guys who dropped out eventually. These were
very suave, sophisticated, social—and I don’t mean in the high sense. I
mean these guys knew their way around. We had rules. When we had a dance with a girls’ house plan, we couldn’t
stick with one girl. We had to circulate. There was no ownership of
these girls. So they had these rules. We were really hip and suave, and
we knew our way around. And we met a lot of fine Italian girls, so I
decided I’d like to be able to speak to the Italian girls in Italian. So
at work we had an Italian guy, and he taught me some high Italian. So he
said, “When you first meet an Italian girl, you say, ‘tanto piacere,’
which means ‘There’s much pleasure in meeting you.’ Then you say, ‘vuoi
ballare?’ which means ‘You want to dance?’ And then, ‘Si balla bene,’
which means ‘You dance well,’ and you go from there.” So we went to this dance and I met this very beautiful, clearly cultured
Italian girl. I went over to her with my Italian. She laughed at me,
because what I had said, I didn’t say, “tanto piacere.” I said, “tanto
pisciare,” which means “You urinate a lot.” Well, that was a good
beginning, and we got along just fine. We had a good time. We went on these escapades on the weekends. We’d
raise hell. But during that time, at one of these dances a girl I knew
came up to me and said, “This young lady would like to meet you.” And
she introduced this woman named Gail to me. She was in high school at
the time. We started dating, and I was able to marry her. I married her
while I was in evening session at night because I was earning money. By the way, when I was earning that money, it all went into the house. I
started earning $35 a week. Then eventually I started earning—at the
end, it was $100 a week. So it was $5,000 a year, which was a big
deal. In November 1954, I married Gail in a civil ceremony so that we could
apply to live in a project in New York City. (You had to be married.)
Then she went home to her home, I went home to my home, and in January
of ’55, we had our formal wedding. Then we took an apartment. So in some
sense, I realize now that was the only way I could justify leaving my
folks’ apartment and stepping out of that environment where I was kind
of a slave, if you will. It was this rigid working, going to school, all
the money piping in. So we took an apartment, and from January of ’55, I put Gail through
high school and college. She went up to Jackson College, which is part
of Tufts University, came back, went to Hunter College, and I kept
working at night putting her through college, and myself. So I got
married quite young. I was twenty years old when I got married, which
was too young.
- FIDLER
- Maybe you can tell me about how you got into MIT.
- KLEINROCK
- Sure. So I was at City College, decided I did want to get a master’s
degree in electrical engineering. It was a worthwhile thing to do. I was
thinking about different graduate schools, recognizing, again, I need to
get a very strong scholarship to go there. I now had a wife.
[laughs] I heard that on a certain day, there was this representative from MIT.
In fact, MIT Lincoln Laboratory was going to come by and describe this
wonderful fellowship scholarship program at four o’clock one afternoon.
So I spoke to my boss and I said, “Look, I need to leave work a little
early this day. I’ll make it up during the week.” He said, “Fine.”So I got up there for the presentation, and this guy from MIT Lincoln Lab
described what they call their Staff Associate program. It’s a fantastic
program whereby you become a student at MIT as a research assistant, you
spend your summers at Lincoln Laboratory as a researcher, you spend one
semester as a full-time student, and you get a master’s degree. They’ll
pay your full tuition, you earn some money as a research assistant, as a
summer employee, and you’ll go to MIT, but you have apply for this. This
was sounding like nirvana to me. What could be better? MIT, the best
place to go to. So the gentleman at that lecture said, “And if you want an application,
go see the professor in the back of the room when my lecture is
over.”The lecture was over, I went back there, I went to this professor and I
said, “I’d like an application.” He was an electrical engineering
professor. He looked at me and says, “I don’t recognize you. Who are
you?”So I told him my name, and I said, “I go to evening session.” He said, “Evening session? Get outta here. You can’t have an
application.” I said, “What are you talking about?” He says, “Get outta here.” I couldn’t believe the attitude. So I wrote away to MIT Lincoln Lab, I got the application, I applied,
and, happily, I was the only one accepted because I was first in my
class in day and evening, the credentials were good, and I presented
well. So I managed to get that application. By the way, I won’t tell you
the name of the professor, but his name basically was appropriate for
that personality he exposed in that little interaction.
- FIDLER
- Can you tell me about your time at the Servomechanisms Lab?
- KLEINROCK
- Sure. Oh, by the way, true to my idiosyncrasy of trying to keep my
options open, even when I was accepted at MIT, I wasn’t 100 percent
convinced that I should go there, which was idiocy when you think about
it. I was thinking, well, where else might I go, what else should I do,
should I really get a master’s degree. And my dad, at this point, was
very clear. He said, “Don’t think twice about it. That’s the place to
go.” So bless him. By now he was working as a clerk again. He had gotten
enough health back, so he was working at that point. So I had a choice of which laboratory I go to. I got up there in January
1957 and started working at the Servomechanisms Lab. The reason I chose
that laboratory is because I had taken a course in servomechanisms at
City College and I was fascinated by these automatic control systems.
They were wonderful systems which, with the feedback loops and the
ability to adapt to behavior, were fascinating. And along with
servomechanisms, you learn a lot about analog computers, because in some
sense a servomechanism exploits the abilities of analog computers. So on my application to MIT, I said, “I’m interested in servomechanisms
and analog computers,” explicitly leaving out digital computers. That
was before I finished City College. I had one more semester to go, and
the last semester after I made that application, I took a great course
on digital circuits, a book called Pulse and Digital Circuits by Millman
and Taub. I still have that book. It was a great book, and one of those
authors was teaching at City College as well. So I went to the Servomechanisms Lab, and the first group I was with was
in a group doing a digital flight simulator, using digital circuits to
simulate flight in an aircraft, a pilot in an aircraft. The group I
happened to be working with were all from Texas. The students were all
from Texas. Now, I figure when I go up to MIT, I’m going to pick up a
Boston accent. These guys all had a southern drawl, and they kept
talking about things I couldn’t understand, like they kept talking
something called a “dowd.” Turns out a “dowd” is a diode. [laughs] So
little by little, I picked up a Texas accent, Boston accent, and my good
New York accent.I did some very interesting work there. Now, here I was in an
environment, all graduate students taking their courses, learning to
work with them, sharing the experience with them of going to class,
doing our research, finding problems. It was a wonderful experience.
These were really smart guys. MIT typically took the top student of
almost any school, and they really filtered in some great people. I had a supervisor of the laboratory named Frank Reintjes. I’m trying to
remember the head of my digital flight simulator group. I think it was
Connor. And I had an advisor whose name was Gardner, Murray Gardner,
Professor Murray Gardner, and Gardner was also teaching the first course
I took on linear and transient systems. I was told by my people at the
Servomechanisms Lab that it would behoove me to do well in this course
because this is the course where they separated the men from the boys.
And guess who was teaching this class. Murray Gardner. He was the author
of the book. So I strolled into class and I did my work. Took the midterm exam. A
grade came back, and I got a 50. I had never seen anything south of 95
for years. This was a terrible blow to me. The other students did okay.
I eventually got the score up to 70. It was totally unacceptable. I
remember driving back and forth to MIT those days and shouting, “Fifty?
This is unacceptable! You can’t do this!” And I realized a couple of things. First of all, I realized that my
study habits were not sufficient to deal with MIT and the students with
whom I was competing. In addition, my education at City College, as good
as it was, didn’t nail down certain understanding of the things he was
teaching which would have helped me in that class. So I decided I better
buckle down, and I worked like a dog, and I got an A in the course. So I like to refer to this as a wakeup call. It was a wakeup call, and
you have to realize that with a wakeup call, you do one of two things:
either you get crushed and you go away, or you rise to the occasion and
you change what you’re doing to correct and make sure that doesn’t
happen again. So I did the latter, of course, and I got an A in the
course, and that was a rather interesting thing. So that wakeup call made me realize that this collection of people with
whom I was sharing courses was a very special group. These kids were
smart and they were driven, and they were weird, nerds upon nerds, but
brilliant. Each one had a different kind of approach to work, nothing
standard, so you really had to stay alert and do your best and do the
work, do the hard work. So I took a bunch of courses with these kids,
began to know them very well, and I was doing my research.After that first semester, in the spring of 1957, I spent that next
summer at MIT Lincoln Laboratory, as was supposed to be the case,
working in a group which was involved with a transistorized computer
called the TX-2, with a bunch of brilliant engineers who were there
full-time and with some summer students as well. This group I was with
was, again, a great privilege because these guys were powerhouses. They
had designed these machines. In fact, this group had conceived the
machine, designed it, built the hardware, wrote the software, did the
compilers, wrote the applications. This machine was part of them, and it
was an innovative machine. The TX-2 happened to be a 36-bit word
machine, which you could break into four 9-bit segments or one 18 and
two 9s or two 18s or one 36. You could do parallel computation. The
instruction set was unbelievable. It was one of the first transistorized
computers.The kind of work going on, it was headed by a fellow named Wes Clark,
Wesley Clark, and he had a bunch of people working under him, and under
those people were people like myself, students. One of the guys I was
working with was a fellow named Jack Raffel. If I’m not mistaken, he had
also come out of Bronx Science, actually, and he was very much
interested in thin magnetic films for storage.But even before I worked for him, now as I recall, I actually started
working for Ken Olsen. Now, Ken Olsen you probably recognize as the
gentleman who founded Digital Equipment Corporation. I started working
for him that summer of ’57. That’s right. He had me studying the
behavior of gamma rays on transistors because he was concerned about the
impact that those gamma rays would have on the functioning of these
transistors when they were part of a computer circuit. So I did that,
and I got some information that helped. Then he had me design a very
simple device, a digital device called a pulse delay amplifier. What you
do is you have a little circuit and you dial in a number like 100
milliseconds, and this box, when you hit the trigger, nothing comes out
until 100 milliseconds later, a then a little pulse comes out. So you
could decide when you wanted a pulse to come out to use as a test for
many other circuits. So I built this thing for him and it performed very well. When I brought
it to him, he wanted it to perform better. He was a wonderful
supervisor. He said, “Len, try a little more. Here, try this.” He was
very encouraging in a soft way and, sure enough, I made significant
improvements under his tutelage. That summer, as I was working for him,
he said, “Len, I’m going to form a company, and I’m taking some of the
people from here with me, some of the key engineers, and I’d like you to
join me,” because he liked what I did for him.I said, “Ken, I can’t do that. I’m on this wonderful program. I’m going
to get my master’s degree. My graduate work is more important.” So it
was fine. So he went off and formed DEC that summer, went to Maynard and he went
the Barn, as they called it, where they formed this group. He took
people like Ben Gurley and Tom Stockebrand and a few others, and formed
Digital Equipment Corporation. One of the first things they built was
this pulse delay amplifier, because they started out selling modules.
They sold flip-flops, they sold amplifiers, they sold a variety of
things, and eventually they sold computers. But DEC, it turns out, has a rather checkered early history because Ben
Gurley, one of the brilliant engineers who designed the systems for
them, got shot to death in his breakfast room while having breakfast
with his family, shortly after DEC was formed, by a disgruntled employee
at DEC. Another fellow, who was one of their technicians, when he came
out of his home, on his way to his garage, somebody came over and shot
him in each arm and each leg and then left him. He survived and he never
told who it was. There was intrigue going on there. These are the way
these young companies would get into it—it was on Route 128, the famous
Route 128 around Boston. That could have been the Silicon Valley had
somehow the environment fostered the kind of environment that they did
at Stanford. At any rate, so I continued on. Then I started working back at MIT as a
student, and I was looking for a thesis, a master’s thesis, and some of
the work I was doing at Lincoln Laboratory under Jack Raffel, which is
after Ken left, was on thin magnetic films. A thin magnetic film is a
very small dot which you deposit in a vacuum chamber, made out of
magnetic material, and it can magnetize one direction or the other,
which represents a bit, either a one or a zero. You read it out the way
you read out from ferrite core memory. You flip it and you see which way
the magnetism goes. So it was a pretty standard but a very compact
memory. However, there was another way to read it, and that was by shining
polarized light on this magnetic dot. If the dot was magnetized in one
direction, the interaction between the polarized light and the
magnetization would cause the polarization to shift in one direction or
the other, depending if it was a one or a zero. So here was a way to
optically read out the contents of memory without disturbing it. It was
a passive readout. So this was the topic, on Jack Raffel’s
recommendation, that I do my dissertation.So I started studying this. There was a well-known effect called the
Kerr, K-e-r-r, Kerr magneto-optic effect. By the way, there was a
well-known, something called the Faraday effect, where if you shine
polarized light through a magnetic field, that’ll also rotate the
polarization, depending upon which direction the magnetic field is. This
was reflecting as opposed to transmission, and the effect was small. So
the challenge was to somehow do the physics to add something to the
magnetic film which would enhance the shift of the polarization to be a
more dramatic readout. Now, this was not quite electrical engineering. It was more physics,
depositing thin magnetic films, putting layers on top of them, but it
was fun because I could create my own apparatus. You needed to basically
shine the light through a collimator onto the magnetic film, and look at
it through a polarized filter, and there was optics and telescopes and
some engineering, and I made my own little apparatus. I was used to
building things from Photobell, from a kid with the radios, and this was
right up my alley. So I built this magnificent structure—you can see it
in my master’s thesis—and I did find a way to enhance the effect. But while I was at it, I couldn’t help myself, and I was surrounded by
my other classmates who also work in a variety of things, some on
magnetics, some on electronics, and a lot of the work we were all doing
was trying to add some mathematics and some logic as opposed to just
physics to these problems. So I decided to see what would happen if you
reflected the light beam off one magnetic film, and it bounces up and
bounces off another magnetic film in a sequence. What would be the
resulting magnetization, what would it mean? And I created a logic,
basically a digital logic, where you could do multiplication, addition.
It was a wonderful logic. It was part of my dissertation, my master’s
thesis, but the effect was not really big enough to make this practical.
I did enhance it, I did the mathematics and the logic, and I handed this
dissertation in. My supervisor was the head of the laboratory, Servomechanisms
Laboratory, this fellow Frank Reintjes. Frank Reintjes was one of the
greats of the era of World War II when MIT had something called the
Radiation Lab Series, and they created these wonderful books and
technology all about radar for World War II. So he was a well-known man.
He was very impressed with what I did. So I went through the sequence. By the way, the sequence for this staff
associate scholarship was you’re at school for a semester, research
assistant my first semester. That summer, you’re a full-time employee.
Next semester, you’re research assistant. Next semester, you’re
full-time student. Next summer, you’re full-time employee at Lincoln
Lab. And now you’ve gone through three semesters and two summers, and
you have one more semester to go, but you should be done with your
thesis by then and have only one course left, and that course you’re
going to take by commuting from Lincoln Laboratory to MIT, which was a
twenty-mile bus ride away. They had a shuttle going back and forth. So I was all set to do that that summer of ’58. And if you realize what
was going on, when you finish this program, you get hired by Lincoln
Laboratory in a fine research position, as a research engineer. So my
wife and I, we were married—yes, I told you. We decided, “Look, let’s
start a family. We’ll have our first child in the summer of ’58.” Of
course, then I’ll be full-time at Lincoln Lab, I’ll be commuting, having
a full-time salary, ready to go. So my son, Martin, was born in August
of 1958, August 21st, 1958. However, Frank Reintjes had other plans for me. He said, “Len, you did a
really impressive master’s thesis. I want you to pursue that master’s
thesis and make it into a Ph.D. dissertation, get a Ph.D. out of
this.” I said, “No, I can’t. I’m scheduled to be at Lincoln Laboratory. My son
is born. I want to get the full-time salary. I want to be an engineer. I
have my life plan set.” He kept twisting my arm and twisting my arm, and
finally I said, “Look, if I’m going to do this, I want to do something
that’s going to have significance, not just piddle around with
something,” because, you know, I enjoyed my master’s thesis, but it
wasn’t earth-shattering. I did my first paper on that, by the way. I did
the paper in Chicago, but that’s another story. I’ll tell you later if
you want. So I said, “Look, I’ll do this.” I spoke to my wife, Gail, and we
agreed, “Okay, let’s do it, but you’ll need another Staff Associate
program to support you.” And sure enough, Lincoln was willing to extend
my master’s to a Ph.D. It was the first time they took a master’s Staff
Associate and extended it out to a Ph.D. So I was very honored to have
that, and the money was good enough. And I said, “If I’m going to do
this, I want to look for the best professor at MIT. I want to do
something which is challenging, fun, and has impact.” So that was sort
of the deal I made with myself. So that summer I decided who’s the best professor I know, and the best
professor at MIT that I knew, the most exciting was Claude Shannon.
Claude Shannon had come to MIT in 1958. He had come off a career at Bell
Labs, having created this magnificent field called Information Theory
and Coding Theory. The man was world-famous. He’s a brilliant mind. I
had taken his classes. He was now at MIT, and I watched the way his mind
worked and the way he worked. He had this wonderful mix of great
intuition, physical engineering intuition about objects, and mathematic
prowess, and he could merge the two. So he could look at a system,
decide what are the underlying principles, use the mathematics to
extract them, exploit them, and then implement them. And he was fast,
one of these really brilliant guys, and he had proved himself with this
classic paper he produced in 1948, which became the entire field of
Information Theory and Coding Theory. He and Norbert Wiener basically
were contemporaries, and Norbert Wiener was also at MIT. So I was surrounded by really brilliant guys. I said, “Shannon’s the guy
I want to work for.” So I called him up that summer of ’58, and I said,
“I’d like to work for you on my Ph.D.” So he said, “Look, why don’t you come out to my house this weekend, and
we’ll talk.” At this point, I was working with Jack Raffel that second summer, and I
told Jack and a number of my classmates and colleagues there, “I’m going
to visit Claude Shannon this weekend.” They said, “Well, that’s terrific.”And Shannon had told me do I mind that he has a houseguest while I come
to visit. I said, “Of course not.” Here I am, this miniscule graduate
student talking to the giant, Claude Shannon. He said, “I have John
Pierce here.” So I told my Jack and my other friend, “Jan Peerce is with Shannon.”
Now, Jan Peerce, as you know, is a famous opera singer. They all said,
“Well, look, why don’t you ask Jan Peerce to sing ‘The Bluebird of
Happiness’ or something,” as a joke. So I visited Shannon that weekend. The man I saw as his houseguest did
not look like Jan Peerce. Jan Peerce was a somewhat rotund person, and
this gentleman was a thin, scrawny-looking guy. And I realized later it
was John Pierce. John Pierce was a colleague of Shannon’s from Bell
Telephone Labs, and John was a curmudgeon and a cynic. And here I was,
summer of ’58, right after my son was born, I’m bubbling over with life
and joy and a new child, and it came up in our conversation, and Pierce
says, “Bah. Children. They’re selfish. They’re horrible. All they do is
disappoint you and exploit you.” And I couldn’t believe this man had
this attitude, and here I am just optimistic, so it was a clash of world
view, certainly. But I spoke with Shannon, and we were out on his porch, I remember very
well. I’m sitting on a chair facing Shannon, and we’re chatting, and
every so often his eyes glance above my head. So finally I wonder what’s
behind my head. So I spun around, and his son was on a hammock, and the
hammock was swinging, the rod of the hammock was swinging within inches
of my skull, and had I’d leaned back, I would have been bopped on the
head pretty hard, and Shannon, I think, was just waiting for that to
happen. [laughs] He was a bit of a—what shall I say—a mischievous
guy. We talked about a lot of things. He told me about the automatic
lawnmower system he had. He had wired his lawn with an underground wire,
and the lawnmower would follow the wire. It was one of the first robotic
systems. He told me about this school bus he had converted into a family
bus. It was sort of one of the first microbuses that people go camping
in. And he told me about the project he’s working on now. The thing that was
really exciting him had nothing to do with Information Theory. He was
working on a chess-playing program with another professor at MIT, John
McCarthy, who is one of the great artificial-intelligence people.
Shannon was interested in getting a program to play chess, basically the
first chess-playing program, because Shannon was a chess enthusiast. He
had these wonderful chess games. He would buy any automatic chess game
where the pieces would try to move automatically.So I said, “I’d love to work on it,” because this was a challenging
situation, and I wanted to work with Shannon. So he took me on as a
research assistant, and the deal was this. John McCarthy and his student
were going to create what’s called a legal move generator. Given the
position of the board, what are the possible legal moves? Then Shannon
would come along with me, but mostly Shannon, and decide what’s the
right move to make. And we wanted to study the middle game, because the
beginning of the game is sort of pat, and end game is sort of strict,
but the middle game was where the really interesting strategy takes
place. So we wanted to build some interesting rules into this program. So
Shannon handed me this book by Fred Reinfeld called 1001 Winning Chess
Sacrifices and Combinations, I think it was called. On every page there
was a position shown and it would say, “Black has a brilliant sequence.
Find it.” Next page, “White has a brilliant sequence.” And you’d look at
this. It was really hard to solve these things, but in the back of the
book were the answers.So Shannon said, “Go to the back of the book, look at the answers, and
find out the following: what is the most common first move of a
brilliant sequence. Because I want to know that move so I can build it
into the algorithms and the heuristics of my stress strategy. Find out
what is the second most common first move of a brilliant sequence,
what’s the most common second move of a brilliant sequence, etc.,
classify them, get those, and we’ll make our strategy based on
that.” Well, I did that. It was easy to do, and it turns out, by the way, that
the most common first move of a brilliant sequence is check. You totally
constrain the opponent. There’s only a few moves he can make, and then
you can take a long sequence in your mind. And I think the second most
common move was a capture. Again, you’re nailing the guy. So we built
this, and we created the first chess-playing program. I worked on that for a while, but I decided this is not really what I’m
particularly good at. And I was taking more courses at the time, and I
was getting ready to take the qualifying exam for the Ph.D., which you
have to pass before you’re allowed formally into the Ph.D. program. I
took Shannon’s courses. I had taken some. I took more. I looked around me and I saw so many of my classmates were working
on—guess what—Information Theory and Coding Theory, which was Shannon’s
forte, and why many of the students had come to electrical engineering.
And by the way, I was no longer at the Servomechanisms Lab. I was now at
the Research Lab for Electronics, RLE, with a different group of
students. And by the way, one of my classmates, who was also on the
Lincoln Laboratory Ph.D. Staff Associate program—he hadn’t got to the
master’s, but was pursuing the Ph.D., was Larry Roberts, and another one
was Ivan Sutherland. These names will come up later as we go along. So I looked around. I saw most of my classmates were working on
Information Theory and Coding Theory, and I also saw the problems they
were working on. We were in a bullpen of graduate students all mixed
together. I would talk to all of them and learn what they were doing,
and the problems they were working on seemed to me to be really hard
problems. And in some sense, Shannon had done most of the critical and
important and fundamental work, and what was left was not only hard,
but, in my mind, was of relatively of little significance. And I said,
“That’s not what I signed up for. I signed up to work on something which
would be challenging, fun, new, and would have real impact.” So I decided not to pursue that avenue, and I decided that the
chess-playing was not really where I shone or really where there’s the
kind of thing that I wanted to do. But having spent time at MIT Lincoln
Laboratory, surrounded by computers, and here at MIT itself, surrounded
by computers, I recognized that sooner or later, these computers are
going to have to talk to each other. These were isolated computers doing
their wonderful work and their wonderful projects, but each one had
projects of interest. They probably want to communicate. You may want to
go from one computer to another and find out what they were doing when
you’re on the first computer. So I said to myself, well, how are these computers going to communicate?
What was going to be the underlying networking technology? And there was
no networking technology, no communication technology which would allow
that to happen. I thought, “Oh, this is interesting. Here is a problem
that I can see is important, nobody’s working on it. So what kind of
network would allow them to communicate?” Well, there was a network around at the time; it was called a telephone
network. Telephone networks were everywhere, all over the world, but I
realized that a telephone network technology would be inadequate for
connecting computers together, because in a telephone network, when
you’re sending voice, you’re talking most of the time, and the way you
connect is you dedicate a link from you through a network of links to
the person you’re trying talk to, and that sequence of links is
dedicated to your conversation. On the other hand—and you’re talking most of the time. You’re silent
maybe a third of the time, which is a small amount. When computers talk
to each other, I noticed, you sit at a computer, you type a few
characters, then you scratch your head or you pause or you think. A
little while later, you type a few more characters and maybe something
comes back. Most of the time, you’re sending nothing, so there’s no way
you could dedicate a sequence of expensive communication links, data
communication links, for this sporadic, bursty communications. In fact, if you characterize what kind of communications computers
engage in, it’s what’s called bursty communications; namely, they almost
never want to talk to each other. When they do, they want immediate
access, but they don’t warn you when they want, and they’re going to
send a tiny bit of data. So how can you possibly have a network which
would allow that crazy kind of demand? I won’t tell you when I want it,
I almost never need it, I only want to send a tiny bit, but when I want
it, I want it immediately. What do we do? Certainly the telephone
network won’t solve that. So I said the only way to do that is to have a highly dynamic adaptive
system which provides communications for the bursts only when they’re
there, but doesn’t dedicate the path to that communication, but allows
other people to use the resources when you’re not using them, some kind
of a shared dynamic, adaptive, demand access system. So I said, well,
how can I possibly create a network which would deal with that? I seem
to understand it’s got to be highly dynamic, highly adaptive, but coming
out of the MIT tradition, not only can I conceive of such a network, but
I have to be able to predict how it’s going to perform, I have to make a
model of that network, and then I’ve got to try to design it and maybe
optimize it, which means I need a fundamental model of what’s going on.
- FIDLER
- And we’re speaking now about your Ph.D. proposal.
- KLEINROCK
- I’m thinking about putting together a Ph.D. proposal. You’re exactly
right.
- FIDLER
- And before we go further with that, is there anything else to wrap up
with the influence of Shannon and his prior work on your thinking about
these problems?
- KLEINROCK
- Absolutely. Absolutely. Shannon was my role model then, and he still is.
He had a way of looking at systems, as I said, merging his mathematical
excellence with his intuitive understanding. You’d walk in this man’s
office, he’d have a differential gear in one hand and a Swiss army knife
in the other, unscrewing it. This mathematical genius was physically
involved with everything he did.And the work he did on Information Theory took advantage of a very
important phenomenon in statistical systems, and the phenomenon is this,
that if you let nature take a good crack at you, she will expose her
average behavior if you let her work on you long enough. So, for
example, if you toss a coin, a fair coin, half the times it comes up
heads, half the times it comes up tails. If you toss it ten times—this
is Shannon recognizing this—toss it ten times and it comes up six times
heads and four times tails, is that an unusual outcome? The answer is
no. There’s a reasonable probability that could happen. You expect five
and five, it’s the most likely situation, but six and four will happen.
Suppose you tossed it a million times, and it came up six in a thousand
heads and four in a thousand tails. Is that reasonable? The answer,
absolutely not. That cannot happen. With infinitesimal vanishing
probability, it’s going to be very close. With high probability, it’s
half heads and half tails. In other words, if you toss it enough times,
this average behavior of half heads and half tails will expose itself.
It’s a kind of emergent property. Let nature run an experiment many
times, she will guarantee to show you her average behavior. And there’s
a name for this; it’s called the law of large numbers.So I knew that in systems if you mix enough things together or let them
behave long enough or have enough numbers, there will be emergent
properties that you won’t see if you have small systems or small trials.
So I said to myself, if I’m going to study these networks, I want to
study large networks where these emergent properties will come out and
maybe expose some principles of network behavior. In fact, when I
started working on this, I said these have to be large networks I’m
going to study. In fact, my thesis proposal, the original title of the
proposal was “Information Flow in Large Communication Nets.” And the
reason that word “large” is in there, for what I just said, to expose
the emergent behavior.That’s Shannon’s influence right there. He said, “Give me long sequences
of code words, and I can predict the way they’re going to behave, and
I’ll put in just the right amount of protection to combat the noise,”
said Shannon, “and I’ll be able to undo any errors that come in. If you
give me the right amount statistically, I can predict what’s going to
happen.” Large systems are predictable, even though they’re random and
statistical. So that was a very strong influence on how I decided to
think about this problem.
- FIDLER
- So in the process of deciding to and how to look at networks, we’ve got
the ideas of Claude Shannon and we’ve got this context of being around
computers and thinking that maybe they’re going to need to talk to each
other. Is there anything else to add to that?
- KLEINROCK
- Yes, definitely. I needed an approach. I needed some tool which would
allow me to take this system whereby these bursty demands are coming at
you all over the place and they’re trying to use the system somehow. How
can I predict the way these unpredictable demands, unpredictable in
length, unpredictable as to when they’re going to make their requests,
how do they behave? What tool is there out there that I can use?And it turns out there is a statistical tool out there, and I was lucky
enough to find it, called queueing theory. It studies the way queues
behave. What is a queue? Think of a queue at a bakery store. There are
clerks behind the bakery store counter, and customers arrive when? Well,
you don’t know. Unpredictable. And when they come in and they occupy a
clerk, how long will they occupy the clerk? You don’t know. But queueing
theory studies these unpredictable statistical systems and shows you how
to predict the behavior; how long will the queues form; how long do the
buffers have to be; how much storage space do you need to store
customers in the store; how many servers do you need; with what
efficiency will they work; what’s the throughput of the system? So these metrics that I just mentioned—throughput, buffering, delay,
efficiency—these are exactly the terms you want to be able to solve when
you have these bursty demands of computers talking to each other. How
many messages per second can I send through? How large must I make those
communication channels? How many clerks do I need, if you will? How many
buffers do I need to store the messages while they’re waiting to get
served? Queueing theory provides exactly the right metrics to study and
provides a mathematical structure to pose and answer these
questions. So I said—ah!—I’ve got a mathematical approach. I’ve got a need. I’ve
got a new problem. There’s a lot of low-hanging fruit here. It’s clear
this will have an impact. And by the way, when I started to look into
the queueing theory, it turns out there had been some work just in 1957
by a fellow named James Jackson, who happened to be a professor at UCLA,
and he had studied a job-shop model, where work would go from one
station to another station to another station in the job shop. First you
cut it; then you saw it; then you drill it; then you mill it; then you
put it together; then you paint it. This flow of jobs goes from hop,
hop, hop, from storage place to storage place. That sounded to me like
messages passing through a sequence of nodes on their way to the
destination. He had some mathematics in queueing theory. So I had a real
approach with an example of how to use that mathematics and a problem
which is important, was current, low-hanging fruit, and one that people
hadn’t studied. Boy, this was right up my alley—just what I was looking
for. So there I was, and I started looking at that problem.
- FIDLER
- And in what other lineages do you place your work?
- KLEINROCK
- So besides Shannon and besides Jackson’s work, which influenced me very
strongly, it was a very important part of my own proposal, my thesis
proposal for the doctorate, the whole concept of queueing theory is
there’s a long tradition of queueing theory which is wonderful
stuff.Now, there is a father of queueing theory. His name is A.K. Erlang, and
did his work in the early part of the twentieth century. He was an
engineer in the Copenhagen Telephone Exchange, an engineer, just like
Shannon, with wonderful mathematical and intuition capability.In telephone systems you have the problem that you have a building, say,
a commercial building that needs telephones. Population may be fifty
people, each with a telephone, but how many live telephone lines do you
need leaving that building? Because most of the time people are not on
the telephone.So he needed a way to calculate how many lines should they supply to a
building or a facility. So he had to make a model, and he created
basically a queueing theory model, because when you place a telephone
call, it’s like an arrival. If somebody’s using a telephone line that
you want and you have to wait, but that’s not good, okay, so he wanted
to make sure there was no busyness, no loss. When somebody comes in and
they can’t get a telephone, they’re lost. He said, “I want to provide
enough lines so that almost surely whenever somebody wants a line, it’s
available to them.” So he developed a model, an intuitive model, in which he wrote down
equations of how these systems would behave. He studied both lossless
systems, where there’s always a line, or even systems where you queue up
and wait. He created queueing theory and he created a mathematics using
a kind of engineering approach to solving the mathematics, and he turned
out to be right on. His mathematics was correct. It was not rigorously
derived.It wasn’t until years later in 1937, when William Feller, the great
probabilist, created something called the birth-death process, where he
was able to show that Erlang’s work could be put on a firm mathematical
basis, and his equations were correct. So here again we have an engineer
solving an engineering problem using his mathematical view of the world
to write down equations, solve them, and come up with numbers which
allow you to build systems. His approach to life was very influential on
me, because I always consider myself an engineer, not a mathematician, a
physicist rather than a mathematician, physical stuff that you build and
make work, and you make it work efficiently. So after Erlang, there were many, many others who produced a theory. I
happen to have this book here, which is the Life and Works of A.K.
Erlang, which is a wonderful book. It’s a very precious book. It’s very
old. It was written in 1948, but it begins the work of Erlang in 1917, I
believe. At any rate, this man, his approach to life matched my view of
life, which was not to come at it from the mathematics point of view
only and leave it there as a sterile set of equations, but to come at it
from a problem that needs solving, bring together the tools you need, be
it the engineering tools, be it the mathematical tools, solve the
problem, and then implement it. Shannon did the same thing. And I
admired that so much in both of these men, but I didn’t knew Erlang, but
I certainly knew Shannon. So the answer to your question is Erlang, the
father of queueing theory, was very influential in my thinking.
- FIDLER
- Now perhaps we can move on in the context of your dissertation, even, to
the relationship between queueing theory and packet switching.
- KLEINROCK
- Okay. So the underlying concept that allows computers to talk to each
other is the fact that I shouldn’t give you a resource, an expensive
resource as a communications path or link, until you need it, and only
as long as you need it. So the idea of this network I was trying to
create had to provide communication links, data links, which were
dynamically allocated to messages only when they were there to be
transmitted. I’ll give you a homely example, by the way. Do you own an
airplane?
- FIDLER
- No.
- KLEINROCK
- But you need a seat on a jet plane every so often.
- FIDLER
- Correct.
- KLEINROCK
- But you can’t afford it. It’s too expensive.
- FIDLER
- Let’s go with that.
- KLEINROCK
- You can’t afford to own the airplane.
- FIDLER
- Correct.
- KLEINROCK
- Because if you did, you’d almost never use it. So what do you do? You
only ask for it and get to use it when you need it. You reserve it.
- FIDLER
- Yes.
- KLEINROCK
- It’s allocated to you. It’s yours. Get off the plane, someone else gets
on that seat. That’s exactly what we needed here. These expensive seats
were the communication, the data links in a network. Make them available
and let people grab them when they need them, and if more than one
person tries to grab it at the same time, they wait on a queue. Three
people need it; only one person gets it. So the idea of a queue forming in front of these important links was
natural. These queues are going to form and shrink. People are going to
move up the queue, get served, etc. So the question is, how can we
dynamically allocate these resources? The answer is using the notion of
a queue. You arrive, either it’s there and you use it, or you wait, your
turn comes up, and then you use it. Nobody owns it ahead of time. It’s
not reserved ahead of time.So how can we do this? Well, throw messages into a network, and they go
hop, hop, hop from node to node. When they’re received at a node and
they want to go to the next hop, somehow it’s decided which next link
they should follow. If that link is free, zip, you transmit over it. If
it’s busy, you wait in a queue. So the idea of queues forming in this
network as these messages compete for lines was an important piece. It
was a natural network of queues model, similar to the one that Jackson
had laid out in his job-shop model. But the issues are now what path do you choose in that network, who
controls that path, and what do you do if there’s a very long message in
front of a very short message? You certainly don’t want the short
message to wait a long time for a long message. If somehow you could
reverse it and let the little one go first, he causes a small
disturbance of the big message instead of a large disturbance to the
little message. Well, how are you going to figure out which the short
messages are? Well, it turns out—
- FIDLER
- Is this related to implications of a queue not being reserved ahead of
time?
- KLEINROCK
- Exactly. Nobody owns it. When somebody gets there, if the guy in front
of the first one to use it is going to occupy that channel for a very
long time, that’s not efficient use for the little guys behind. Now, it turns out this exact problem had been faced by the computing
community years before in the context of time-sharing. In a time-sharing
computer, the resource is the processing engine, the CPU. In a network,
the resource is the communication channel. Going back to time-sharing,
they had a problem. Jobs would come in, they want to use the CPU, and
they want to serve the short jobs in front of the long jobs, so the
short jobs don’t have to be delayed a long time. So how do you find out
which the short jobs are? Well, you can ask, but every job is going to
say, “I’m a short job.”So what time-sharing did is to say look, we’re going to ask you to prove
that you’re a short job. We’ll give you the processor, you get a little
slice of time. If you’re a short job, that’ll be enough to get you
through. If not, you say, ah, you lied, put that job in the tail of the
queue, and let another job try to prove he’s small. And so as you go
round and round—this is called round robin—as you go round and round,
the short jobs will be filtered out early on, they’ll expose the fact
that they’re short; they’ll get through. The long jobs are going around
more times, which is fine because they’re being delayed by a small
amount by the little jobs.So the idea to implicitly find out which are the long jobs and which are
the short jobs comes out of this round robin. I said, well, that’s an
interesting thing. Why not incorporate a similar idea in the networking
model? A long message comes in, chop it into fixed-length pieces. Each
of those lengths corresponds to the amount of service they would have
gotten in the time-sharing system. Let those little fixed-length things,
which we now call packets, make their way through the network, so nobody
gets stopped by a big thing; they they only have little packets in front
of them. And your long job, your long message becomes a sequence of
packets which flows through the network, gets collected at the other
end, and shipped out as a message. So in my early work, when I presented my thesis proposal, I said, “Look,
we want to get an analytic model for what a data network looks like.”
Well, a queueing network model is the right one, and we know how to
solve that. We want to decide how to route messages. Well, we’re going
to look at a variety of routing procedures, including fixed routing,
adaptive routing, dynamic routing, so we don’t force all the control
into one node. We’re going to deal with large systems so that they get
the statistically deterministic behavior, and we’re going to look at the
queueing discipline. As messages go hop, hop through the network, I
point out in my thesis proposal, let’s study various queueing
disciplines because that will help us decide on the efficiency by which
these messages get passed through the network. So I started studying that, and very soon after the proposal was
submitted, I recognized that this adaptive routing, this dynamic
routing, this resource sharing, and this idea of using the right
queueing discipline was very important in order to get the short jobs
through quickly, and so I analyzed all these systems and, in fact, I was
able to get an exact analysis for this time-sharing, this round robin
time-sharing system, and I was able to prove mathematically that the
short jobs did do better than the long jobs did. The long jobs did worse
than average, short jobs did better than average, which is exactly what
you wanted in these data networks. And so I included that among the
possibilities as to how to deal with the queues in my thesis. So packet
switching came out of that thinking.
- FIDLER
- And to get there, there was an Independence Assumption that you
introduce.
- KLEINROCK
- Oh, yes. So I set up this analytic model, started working out the
details of its solution, and in order to get an exact solution—oh, by
the way, the other thing I included in my thesis proposal and in my
results was a field of study called network flow theory. Network flow
theory talks about graphs and links and flows through these networks to
determine how much can you push through a network with finite capacity
links.Well, guess what. Shannon, among others, has solved that problem. He came
up with something called the max-flow min-cut theorem, which tells you
exactly how to predict the maximum flow you can get through a network of
finite capacity links. Here’s Shannon again. His fingerprints are all
over this domain. So basically what I did was to set up an analytical model and then try
to solve it exactly. Now, it turns out if you look at the mathematics
and you look at the exact equations, this turns out to be an intractable
model for two reasons: one, the statistical side drives you nuts; and
secondly, the network flow side drives you nuts.So I had to find an easier way, some way to either give up on this
wonderful problem I have created or do something to adjust the model so
I could break through the analytics, and I decided what I’m going to do,
I’m going to make an approximation. I’m going to make an approximation
so the thing which is making the dependence in my statistical model so
difficult and impossible to evaluate is removed and allows you to push
through to a solution. And that particular assumption I called the
Independence Assumption. The problem is that as messages move through the network and packets,
they maintain their same length as they move through, and when they bump
into each other, they cause a dependence in the mathematical equations
which you just can’t solve. We still haven’t solved that problem today.
By making an Independence Assumption, it allows you to remove this
complexity. The Independence Assumption, to be specific, says that every
time a message encounters a new node on its way, hop, hop, hop through
the network, its length is randomly chosen again, independent of what
its original length was. Two comments. One, that is clearly a false assumption. Comment number
two is, does it affect the solution? Does it affect the things you want
to solve for? And what do you want to solve for? You want to solve for
how long it takes messages to get through the network, how many can go
through, what the efficiency is, etc.So I was of the impression that this Independence Assumption, first of
all, would definitely allow me to do the mathematical solution, and I
had a suspicion that it wouldn’t alter the results very much. So I had
the following issue. I can put the mathematics through and get an
answer, but then I have to prove that my assumption is a good
assumption. Now, how are you going to prove it? You can’t build a
network. It would take years and millions of dollars, but you could
simulate a network. So I decided what I’m going to do was to use my knowledge, that at MIT
Lincoln Lab there’s a great computer out there I can use. I can create
this massive simulation of very large networks and simulate them in two
cases, with the assumption, and without the assumption, measure the
things I want, like the response time, the delay, the throughput, and
see if the results are close or not. If the answer is they’re not, I
have a problem. If they’re close, I’ve got a miracle. And so that’s
exactly the path I took. What I did is I was able to solve the mathematics, extract certain
principles from that mathematical solution, which I can talk about in a
moment, and once you have an exact analytical solution, you can then
optimize things like where to put the capacity in the network, how much
to put, how to route things, what kind of topologies to use to minimize
the important things like the response time. I was able to do all that,
but I still had the job of proving that this was an accurate enough
solution. So I decided to write this simulation program and run it on
the TX-2. So while I was doing the analytical work, I also did the programming
work, and this is part of my work at Lincoln Laboratory. So I wrote this
very large simulation program, and, of course, my classmates at the
time, both at MIT and at Lincoln Lab, people like Larry Roberts and Ivan
Sutherland and a bunch of others, were present at the time. Larry wrote
the compiler for the TX-2, which I happened to use. I wrote a 2,500-line
program without ever debugging any of it. This is not a good approach,
by the way. [laughs] And then I set out to debug it all at once, and it
took me four months of grueling work to test and debug that program so
it would run. If I wasn’t able to debug that program, I would not have
gotten a Ph.D., because I couldn’t prove what I needed to prove.It was a worrisome period and, in fact, what I had was access to that
TX-2 for hours at a time in the time-sharing mode, and in it’s the
signup mode. I could get that machine seven hours a night, from midnight
to seven, four days a week, for four months running. Well,
unfortunately, they weren’t contiguous days. It was like a Monday,
Wednesday, Thursday, Saturday. So my sleep habits went to hell. My sleep
habits went to hell in evening session. I hardly slept.So I could manage that, and I spent all that time debugging this program.
I remember one night I was there late at night debugging this thing and,
you know, this is a very expensive computer, a million-dollar computer,
all to myself at night with the printers and the tape drives and the CPU
and the thin-film memory and all the sound and the whistles and the
whispers and the cycles, and you get to know every sound. And by the
time seven a.m. comes around, you are grubby and you’ve got a beard and
your mouth tastes lousy and you’re tired. And in comes one of your
classmates. He’s on the seven a.m. shift, he’s bright, he just had
breakfast, he’s shaved, he smells great, and you curse him for coming.
This was the cycle I was in. So there I am late one night, about three or four a.m., and I’m working.
I’m trying to debug this damn thing, and I hear a sound I’d never heard
before. It was pssst pssst, and I really got frightened, because I
figured something was about to break in this million-dollar machine, and
if it did, it’s my fault. So I’m looking around, I’m looking at the
console. This was an experimental computer, a TX-2 transistorized
experimental computer, and pieces of the machine of the control panel
were missing every so often because they were being repaired. And so my
eyes glanced over various pieces, and they glanced up to an empty hole
in the console, and in that empty space I saw a pair of eyes looking at
me, and it scared the hell of me. And who was it? My good friend, son of
a bitch, Larry Roberts, and he was back there trying to scare me, and he
scared the hell out of me, going pssst pssst.
- FIDLER
- Actually, lets speak a bit about your relationship with Larry Roberts
and also Ivan Sutherland.
- KLEINROCK
- So Larry was on the same program I was. He was working for some
professors, and one of the professors on his committee was Claude
Shannon also. Larry was working on a picture processing program, 3-D
picture processing, trying to find hidden lines in 3-D objects. He and I
were both using the TX-2 for our dissertation, as you see, I was for my
simulation. We shared an office at Lincoln Lab. I met Larry first in
1959 when I started in the Ph.D. program, and we went through the whole
program together. We were very close buddies. We spent a lot of time
talking at MIT and at Lincoln Labs. We went on camping trips together.
We went to see an eclipse, total eclipse of the sun, up in Maine. We
were very close. We shared everything. I knew everything he was doing
for his research. He knew everything I was doing. We shared ideas. We
recommended things to each other. In fact, that was the modus operandi for these graduate students. We
shared offices, and we’d always talk to each other about what we were
working on, independent of the professors. If we’d have a problem
breaking through, we’d get their advice, they’d make a recommendation,
we’d critique what they were doing. We’d teach each other things we knew
that they didn’t know, and everybody was highly cooperative and highly
competitive. It was a wonderful honing operation, where you hone your
skills. You had to prove your mettle to your classmates as well as to
the faculty. These are the people you spend the rest of your
professional career with, in very diverse fields. I could go down a list
of who my officemates there. It’s a rogues’ gallery of the giants of
communications and networking technology.
- FIDLER
- And how did you know Ivan?
- KLEINROCK
- Ivan was also a Staff Associate. He came in from Caltech with his
master’s degree, whereas most of us had gotten our master’s degrees at
MIT, which was very useful to get your degree at MIT, your master’s,
because that prepared you for the qualifying exam to get into the Ph.D.
program, because the problems on that qualifying exam were based on the
material of your master’s degree. However, the questions were not
straightforward. In that year when I was taking that exam, they were all
trick questions, and if you didn’t get the trick, you’d fail that exam.
So you had to be alert, clever, and knowledgeable. So talking about Ivan Sutherland, he came in from Caltech. Instead of
having two years of preparation, he had like two months. I remember
Larry and I and Ivan took it at the same time with some of the guys.
Ivan scored top. Pure raw brilliance, wonderful approach. I really
admire him. He is one of the smartest guys I know, and his name you
don’t hear much about. He’s kept a low profile. He doesn’t give
interviews. Brilliant guy. His personality was that of a nerd. He’d get
on the phone with you and he’d begin talking to you about a mathematics
problem or a science problem instead of saying— I’d usually stop and
say, “Ivan, just a minute. Say hello.” [laughs] You know, a moment of
chat. Now let’s get into it. But he’s since, of course, mellowed quite a
bit. So Larry and Ivan and I and other classmates were very close. In fact,
as a master’s student at the Servomechanisms Lab, I was forced to start
smoking a pipe. It was what you did as a graduate student. I hated
smoking a pipe, but you just had to do it. It was an open bullpen. So
you sort of joined the group, you did what they did, but it was an
unbelievable experience. Anyway, that’s how Larry and—and Larry actually
suggested while I was doing my dissertation that I apply it to
road-traffic theory because road traffic’s also a network of queues.
It’s a queueing network. I never did. Actually, more recently I have
started to do that, but back then we never did that. Larry had on his committee Claude Shannon. Ivan Sutherland’s supervisor,
I believe, was Claude Shannon. Claude Shannon was on my committee. He
wasn’t my supervisor; he was on my committee. So we all shared Claude
Shannon, so we had a kind of common philosophy, approach, tutelage, and
we’d all reinforce each other.
- FIDLER
- Let’s switch gears and talk about your move to UCLA and the beginnings
of the Network Measurement Center, the first ARPANET node.
- KLEINROCK
- Sure. We can get into the UCLA story, but before I do that, I’d like to
say a few words about the research I was able to conduct in that milieu.
- FIDLER
- Great.
- KLEINROCK
- So what I was looking for, as I said, was a way to capture not only the
behavior of these large networks, but what is something fundamental that
emerged out of that study. So just to reiterate, what I did is I created
an analytic model for what a data network should look like. I put in
Independence Assumptions so I could solve that model exactly, and, in
fact, I have an exact equation without the Independence Assumption which
can’t go further.Then I use the Independence Assumption and get a complete solution, and
once you have an analytic solution, there’s a few things you want to do.
One thing you want to do is you want to do an optimization. Now that you
know the behavior exactly, how can you alter the behavior by playing
around with the way you assign the channel capacity, how much capacity
you put in the network, how you do the routing procedure, how you do the
control, how you do the topology, to minimize the critical and important
performance criteria like response time and maximized throughput, and I
did that. But more importantly, having done all that, what principles, what do you
learn from all of this? Are there things that come out? And by the way,
that approach to not only solving but understanding and extracting
principles has been the way in which I’ve conducted my entire
professional career, and, by the way, is the complaint I have about the
way a lot of research is done today. We can get back to that later, if
you’d like. So what was the principle I was able to extract? Well, for one thing, it
became clear, as I said, as I started out, that you want to use demand
access, shared resources, not assigned in a reserved way and wasted, but
dynamically assigned in a dynamic fashion, in an adaptive fashion so
they’re efficiently used. So that resource-sharing was critical. It
turns out that if you share dynamically large resources, you get
enormous benefits in terms of all these metrics, high efficiency, better
response time, more throughput. So a large shared system was a principle
that emerged out of this. The other thing is getting back to the original title of my proposal,
the idea of information flow in large communication nets. I said if I’m
going to design large networks, there’s no way we can put all the
control in one node and have it control tens of thousands of other
nodes. It’s too much of a load on one node. It’s too vulnerable. It
takes too much traffic in and out of that node to collect and distribute
the control information. So it had to be a distributed control system,
which meant that you delegate the authority, the control, so every node
participates in this control function, no one node is doing it all,
which builds in an automatic and free robustness against nodes failing.
If one node fails, no problem. Everybody else is helping in the
control. And in addition, it portends a philosophy which emerged later when I get
to UCLA in the ARPA story. The idea of delegating authority and letting
everyone contribute is a fundamental principle that turned out to emerge
in the management and in the architecture of the network as well. So the
idea of distributed control was important. So, distributed control,
large shared systems, dynamically allocated resources were all critical
principles that I was able to find.I was also able to discover that if you try to concentrate the traffic
into a few large shared systems, large shared networks, then you get the
economies of scale that I referred to before. So, topology information,
routing information, capacity assignment information all played together
to say the same story, statistically: get deterministic behavior out of
it by merging together a lot of things, put a lot of capacity in there,
distribute the control, don’t reserve, but, rather, dynamically allocate
resources were all the principles that emerged. Okay. So now we can get to the UCLA story.
- FIDLER
- And in particular, the factors that led into your decisions to go to
UCLA in the first place. You left the East Coast.
- KLEINROCK
- Sure. Well, it wasn’t an easy leave. There I was at MIT. Lincoln
Laboratory had sent me through my master’s, my Ph.D., in the most
magnificent, benevolent, and effective support, both emotionally,
financially, scientifically, etc., and the environment they provided. So
I was committed in my mind to take my career and spend it at MIT Lincoln
Laboratories, which was a great place to be, lots of things going on
there, very important research, very important, great people. What could
be better? So when I was preparing to do that—in fact, this was in the fall of
1962, I handed my dissertation in in December 1962, but my graduation
was in June of ’63. That’s when the ceremony was, and I had some things
to finish up at MIT in the spring of ’63. I told Lincoln Lab, “I’ll be
joining you, as we’ve all agreed, and I look forward to it.”And here was the approach taken by Lincoln Lab. They said, “Look, we
really want you to work here. There’s a job waiting for you. Terrific.
In fact, we’ll give you a choice. If you want a window office, it’s got
to be shared by two people. If you want an office with no window, you
get your own private office.” I selected the private office; I wanted an
unshared office. But they said, “Look, in fairness to yourself and to
make your future career as productive as possible, we recommend that you
look at other possibilities for where you might go before you commit to
Lincoln Labs,” and they encouraged me to do that, so I did. I decided to look at other places—I went to Bell Labs; they offered me a
position. I went to some of the engineering companies around Boston,
around New York, and out west. I went to Los Angeles and San Francisco,
some of the aerospace companies, the research labs. Hughes had a Malibu
research lab. They offered me a terrific job. In fact, they offered me
$15,000 a year, which was a huge amount of money in those days, because
IBM, Bell Labs, and Lincoln all offered me about the same, which was
$12,500 a year. That was a lot of money in those days—those are huge
salaries. So I went around and made this tour. While I was in San Francisco, it
was recommended that I interview for an academic position to Berkeley.
So I went there, not thinking anything of it. I hadn’t ever considered a
career in academia, but I had done some teaching as a teaching assistant
at MIT, and a teaching assistantship at MIT in those days was you taught
the course, so I really had the students, the whole shebang. So you got
the experience. I found I liked it. I very much enjoyed opening up the
minds of some of these students. So I figured, okay, I’ll interview Berkeley, and I did the interview,
not really planning an academic career on the West Coast. The interview
went very well. One of the faculty that interviewed me and I interviewed
him was a professor named Lotfi Zadeh. He was a great name in those days
in linear systems theory. There’s his book over here, Zadeh Desoer. He
was a giant in the field. He was a tall guy, bald-headed, from one of
the Stans, Turkistan or Azerbaijan or whatever, and a foreboding figure
of a man and personality, very strict, very stern. Took me into his
office and he sat me down in a soft low chair, and he sat on a high
wooden stool and he gazed down at me. The first question out of his
mouth was, “Kleinrock, where do you stand at MIT?" I said I was near the top of the graduate students. Then we went through
the—right away I didn’t like him, with the setup and the question. Went
through the interview and I did fine. Then we finish the interview, and
I was not very comfortable with him. So when we walked out of his
office, I saw he walked down the corridor to the right. I decided to go
down the corridor to the left. And we’re about 100 feet apart, and
suddenly I hear, “Kleinrock!” he shouted down the hallway. It was an
empty hallway. “Where did you stand at City College?” I said, “First!” I spun on my heel and I left. I never heard from Berkeley. Got back, all the other places offered me a
position. Never heard from Berkeley. What happened is that the chairman
of the department, electrical engineering department, stepped down,
another guy stepped in, and they lost my file, I found out later. But I
didn’t think anything of it. That semester, which was the spring of ’63, Zadeh comes on sabbatical to
MIT, electrical engineering. He’s walking down the hall, and I see him.
He comes, and he says, “Kleinrock!” as if I’m his best friend. And we
really hit it off. Really, he’s a very nice guy. Just that first
impression was a little bit off-putting.He thought, not surprisingly, that I wanted an academic position, so he
contacted his good friend here at UCLA, Balakrishnan, and said, “Look,
this is Kleinrock. He’s done this work. I think highly of it. Why don’t
you interview him.” So UCLA set up an interview for me.And while I was at it, I interviewed a few other universities. I
interviewed Stony Brook in New York, I interviewed Rice University,
because a guy named Marty Graham was there, a brilliant guy, and UCLA.
Sure enough, they all offered me a position, but the UCLA position
looked really good. I really liked the department, I liked the school, I
liked Balakrishnan. It was a very interesting place. So here I have this offer from UCLA at about half the salary, like
$7,000 a year, academic year, 3,000 miles away from where I’m living, in
the wild west—in those days, crossing the United States was a big
deal—family on the East Coast. I had a family started. What to do? So I presented this dilemma to Lincoln Lab. I said, “Look, there’s
really this nice offer from UCLA. I’m willing to take that chance, but I
feel an obligation to Lincoln Labs, and I’m not sure I’m going to like
teaching in the first place.”So, god bless them, they said, “Len, take that position at UCLA. If you
don’t like it, you can come back.” Well, with all those options, how
could I refuse? So, sure enough, I took the position in August of 1963.
I drove my family across the country, heading west, wagon train kind of
thing, came to UCLA, and I’ve been here for fifty years. [laughs] [End
of May 20, 2013 interview]
1.2. Session Two (July 20, 2013)
- FIDLER
- So how did you go from having a research focus in computer networks to creating and
running the Network Measurement Center at UCLA?
- KLEINROCK
- So when I arrived here in 1963, I continued to conduct research, both in computer
networks and in other performance-evaluation areas, all of which were analytically and
theoretically designed and focused, and I began to work with Ph.D. students and began to
teach the subject. That was going along just fine, except all the while I was trying to
get a way to convince industry that some of the work I had done would be suitable for
them to take up and implement. The reactions I got from AT&T, as I think we’ve
discussed already, were not the best. They said it wouldn’t work, and even if it did
work, they wanted nothing to do with it. And I called many—I can’t recall, by the way,
if we discussed this in the earlier interview.
- FIDLER
- Not to do with AT&T or efforts with industry. Maybe if you could speak a bit to
that.
- KLEINROCK
- Okay. Sure. So I recall in the early and mid-sixties, there’d be these major computer
conferences, the Spring Joint, I think, and the Fall Joint Computer Conference, and
there’d usually be a plenary session with a group of people discussing networking, and
typically there’d be two constituents: one would be the telephone industry (the
telecommunications industry) and the other would be the computer guys.I remember being up on these panels in front of tens of thousands of people discussing
this, and it would begin where I’d typically say to the telephone guys, “Look, please
give us good data communications, data networks.”And then their answer would be, “Well, the country is a copper mine. It’s full of
telephone wire. So why not use the telephone network?”And my response typically would be, “Well, you don’t understand. It takes you twenty-five
seconds to dial up a call, you charge us a minimum of a three-minute call, and I want to
send a tenth-of-a-second of data.”And their answer typically would be, “Little boy, go away.”And the debate would go back and forth. And as history shows, little boy, we data guys,
went away, and we created what we now call the Internet, which seriously disrupted the
telephone networks’ architecture, and they had to respond to it.But in those days, there was no embracing this new technology by the telephone engineers
or telephone carriers, largely because there was no data to send. There was no revenue
in data, since there was none to send, whereas they were making plenty of money on voice
communications. So from a business point of view and from a short-term point of view,
they were absolutely correct, but their long-range view was lacking. They couldn’t see
that data would dominate soon, and so they chose not to participate at all. So the
reaction of industry was quite negative and somewhat disheartening. Nevertheless, I
continued to pursue the research. I felt it was important to do, to develop technology
as far as we could.Along the way, ARPA gained interest in this. What was happening is in 1957, Sputnik went
up, caught the United States with its pants down. President Eisenhower said, “This will
never happen again,” and he created the Advanced Research Projects Agency in early 1958,
which was an agency within the Department of Defense. The role of that agency was to
bring up the level of expertise in the United States in the—well, we’re now called the
STEM area, Science, Technology, Engineering, and Math, by supporting them in various
ways, mainly in providing funding of research and development in those domains, and so
ARPA was formed and began to do exactly that.In 1962, the computer group was formed under Licklider. It was called the Information
Processing Techniques Office, I-P-T-O, IPTO. Lick was the first director of that office,
and he articulated a vision around that time which involved something he called the
galactic network, where he was talking about man-computer symbiosis and the wonderful
enhancements that could occur if humans and computers did what they do individually best
and collaborated to have this symbiotic relationship, thereby to produce yet further
great advancements, but he had no idea how to do it technically. Little did he know that
I had already published the technology from a mathematical and certain engineering point
of view about how it could be done. But we didn’t know each other at the time, even
though we both had MIT connections at the same time.Two years later, one of my classmates from MIT, Ivan Sutherland, took over as director of
IPTO, and early in his directorship. he came to UCLA and he recognized that we had three
near-identical IBM mainframes. One was in the medical school, one was for the campus,
and one was in the business school, as I recall. He suggested to me why don’t we connect
those three computers together in a small computer network, because he had seen my work
at MIT. We were very close colleagues. He said, “Let’s put them together since that’s a
very nice experimental testbed here.”Well, it never happened, and it never happened because of the political jealousies here
on campus. The three administrative groups were unwilling to yield any control over
their machines, arguing that their machines were being used 100 percent of the time, and
to share them would be deleterious to their efforts. They didn’t anticipate or recognize
the benefits of mutual sharing.Nevertheless, the concept of a network was now present in ARPA’s mind. A couple of years
later in 1966, Bob Taylor became director of IPTO. By then, ARPA had been funding
computer research for quite a number of years and they had established some very
excellent centers of research and development, and these individual centers became
highly individualistic. As an example, at the University of Utah, they developed
excellent graphics processing under Dave Evans and some of the others that were working
with him. And at SRI, Stanford Research Institute, under Doug Engelbart, they’d put
together some very excellent database technology. And over at the University of
Illinois, under people like Slotnick, they created high-performance computing.Every time Bob Taylor approached a new researcher to join this effort and he was prepared
to fund them, the new researcher—these researchers were called principal
investigators—said, “Fine. You want me to do research for you, buy me a computer.”And Bob said, “Fine. We’re happy to buy you a computer.”And the researcher would go on typically and say, “But not only do I want a computer, but
I want to have the same power that each of those other sites has. I want the graphics
from Utah, I want the high performance of Illinois, etc.”Well, Bob recognized he couldn’t fund to the full extent every site with every
capability, so he pointed out to the researchers, “Look, if you were in a network and
you want to do the graphics, you log on to the machine at Utah and run the graphics
there. And you want database access, log on to to SRI,” etc.
- FIDLER
- And just to go back to your time at MIT, this is similar to the situation that you were
in when you thought more about the need to network computers together with the TX. Maybe
you can speak to that.
- KLEINROCK
- Sure. One of the things that motivated me to do this research in the first place when I
was a student at MIT, I recognized that computers eventually would have to talk to each
other, and one of the things that made that clear to me was that there was at MIT
Lincoln Laboratory the TX-2 computer, which was a transistorized computer put together
by the developers at Lincoln Lab.They had earlier developed a transistorized machine called TX-0, the first of the series,
and that machine, TX-0, is now at MIT. And I recognized that there would be an advantage
and the eventual desire for the people who had developed the machine to access the TX-0
from Lincoln Laboratory, and vice versa, those who had now migrated to MIT might want
the greater power of TX-2. So here was a perfect example, in my mind, where it was
natural for people to want to, over some network, connect two computers together, and
that was part of the motivating reason that I had.Well, this was a similar argument that Bob Taylor said. He said, “Look, you want access
to a computer. I’m not going to give you an identical copy of that machine. I’m going to
give you access off of a network.” So Bob Taylor recognized that that was a good reason
to create a network. The germ of the idea was already there by Ivan Sutherland.So here we now had a situation where there was a need, there was a technology that could
be applied, and a significant desire on the part of the funding agency at ARPA, and so,
sure enough, they put forward a plan, and they brought in Larry Roberts as Chief
Scientist in 1966 to manage this full operation. Well, Larry came there. Larry was
another classmate of mine at MIT, and he was a very close classmate. In fact, we shared
an office at Lincoln Laboratory. He was extremely familiar with my work. In fact, he
wrote the compiler for the TX-2, which is the machine I used to simulate my computer
network research. Larry was now willing to take on this task. Actually, he was unwilling
at first. Bob Taylor had to pull some strings to get Larry there from Lincoln. Well,
Larry was working at Lincoln Laboratory at the time.But Larry recognized that this technology that I had published in my dissertation had
proven that this concept of data networks was a reasonable one, that messages and
packets wouldn’t fall on the floor, that there’d be reasonable-size buffers, that the
throughput and the response time were within acceptable limits. So he said, “Look,
there’s theoretical and simulation proof that this works. Now we’re going to make a
physical implementation of it,” in the form of this thing which was eventually going to
be called the ARPA network.So Larry took that on. He put together a group of us to help him specify what this
network would look like. We met. We specified a number of aspects of it. For example, we
had a representative from the time-sharing community there. I believe it was Herb Baskin
from Berkeley. I’m not sure that was the person. But the point was made that if this
network can’t give us a half-second response time, then I can’t get the feel of
time-sharing over this network, because the idea was that somebody at one location will
log on to a remote machine, and that user should feel as if he’s locally connected to
that machine in terms of response time in services.So Herb said, “We need a half-a-second response time.”We said, “Okay, half-a-second response time is what we want for short messages.”Well, it turned out, by the way, by the time we implemented it, we could easily achieve
two-tenths-of-a-second, and so that criteria and that specification were easily met.Wes Clark was making the recommendation that no way do you want to burden these mainframe
computers you’re going to attach to the network with the task of all this communications
data networking. He said, put that in a separate machine alongside the mainframes. So
the idea of this co-processor, which we eventually called an interface message
processor—we now call them routers or packet switches—should be a separate machine. So
that was in the spec as well. We’ll have a separate machine, an unattended machine. It
should not require any human intervention. It should have no rotating parts so they
don’t fail. It should be mechanically and physically strong. So we wanted a hardened
machine to sit in a closet unattended. That spec went in.We also wanted reliability. Now, in those days, the telephone company was claiming very
high reliability for their network of what’s called five nines. Five nines means 99.999
percent of the time, the network will be running properly. What they didn’t tell you at
that spec at the time was that when the network was up and running, it would continue to
run 99.999 percent of the time, but when it went down, they weren’t counting that in
their calculations.So we said, you know, those kinds of five-nine specs are difficult to prove and a little
bit pie-in-the-sky. We wanted a more pragmatic definition of reliability. So what we
said was that the network should be such that if any one piece of it breaks, any one
line or any one switch fails, all the rest of the network should still be able to
communicate. And that turned out to what we call in network flow theory, a two-connected
topology. There’s two independent paths between every pair of host machines on the
network. So that specification was made.
- FIDLER
- Since we’re on the topic here of these Washington, D.C. meetings in 1967, can you speak
maybe more to the construction of the RFQ?
- KLEINROCK
- So the request for quotation had not yet gone out. This was an early meeting where we’re
trying to lay down the specifications. Larry Roberts then took that specification and
created an ARPANET plan, and from that he created this request for quotation, which went
out in 1968.One other point I was going to make, which answers your original question about how we
went from the research I was doing to a Network Measurement Center, was that I
recognized that if this ARPANET was going to be an experimental network, we needed some
way to experiment with it, which means we had to put in some hooks, software and
hardware hooks, to be able to generate artificial traffic, make measurements all across
the network, send those measurements back to a location which eventually we were to call
the Network Measurement Center, to trace packets as they routed through the network, to
measure the levels of traffic and the response times, etc.So I said we have to introduce them to the spec that we’re putting down, the ability to
put in the hooks to run the experiments, to collect the measurement, to generate the
traffic, etc., and that was specified as well. And all of those specifications that we
just talked about went into the request for quotation that Larry put together, and that
went out in 1968.And as a result, it was also decided that since UCLA, we had the expertise in network
technology, that we would be the Network Measurement Center. We’d be the ones to conduct
those experiments, to collect the data, to evaluate them, to match them against what we
expected from a theoretical point of view, and then basically to try to find the limit
of the network performance. The specification was it was our job to try to break the
network, and in so breaking, we would find fault with the network and would, therefore,
find ways to remove those faults and get the network up and running. So that’s the trace
from the theoretical work I came to UCLA to do, to the physical implementation of the
Network Measurement Center here at UCLA.
- FIDLER
- And were there any other challenges or issues involved in establishing the Network
Measurement Center at UCLA?
- KLEINROCK
- Oh, yes. Oh, yes. This was a big issue, because, first of all, we had to get a team
together that had the capability to write the code, to run the measurements, to put
together the hardware, to accept this first switch. The Network Measurement Center was
to be the first node of the network as well, so we had to develop the hardware to match
the IMP-to-host interface, and to come up as the first node and then to talk to the
other nodes as they came up.So I was a young professor at the time. By 1967, I was already an associate professor and
I had my own Ph.D. research students, but I didn’t have a software team and I didn’t
have a hardware team and I didn’t have a staff of secretaries, of administrators, etc.
Happily, one of my colleagues, Professor Gerald Estrin, had such a group that he was
working with, and, in fact, that group was running our time-shared service here at UCLA
on the Scientific Data Systems SIGMA 7, SDS SIGMA 7. And it was pretty clear at that
point that that would be the ideal group to assist me in bringing up this Network
Measurement Center, in becoming the first node, etc., so basically I inherited that
group almost in toto, and to that group I added my own Ph.D. research students, and some
of the other faculty joined with this effort as well to participate in this network
effort, and so the group was created that way.The challenge was now to manage this group that I had not created on my own, and these
were all brilliant people, names that we can discuss later, names that are now known in
the network world, a very creative, opinionated, strong, powerful group of researchers
and software developers, most of whom were graduate students, to work with them and
basically create the esprit de corps and the management philosophy which would allow
them to function properly and to extract the best out of their talents.So here I was, having experienced two types of control and management experience in
coming from my research to this point. The first was that I recognized early on that if
we’re going to deal with a large network,that we can’t allocate all the control to a
single node, and so, as we discussed, we needed a distributed network architecture.
Well, a distributed network architecture means giving control out to many parts of the
network. Each part basically controls a piece of the network. That’s called delegating
from a technical point of view.The second exposure I had to this notion of delegating authority was from ARPA itself.
ARPA was composed of some brilliant program managers there in Washington. Many of them
were themselves recent Ph.D. graduates or researchers themselves who were spending time
at DARPA to promote DARPA’s goal of basically improving the United States STEM
capabilities, and their philosophy was one of also picking the best people, allowing
those people to do what they felt best, without controlling them, by throwing
considerable amounts of money over long-range intervals, without much control, chasing
high-risk, high-payoff goals, which would then allow those researchers, those principal
investigators, to conduct their best work in a free, fluid, and resource-rich
environment.So I said here I was with those two experiences of distributing and delegating authority.
Here I had a group of very talented staff, researchers, graduate students, and my
philosophy was to follow that lead and allow those students, that staff, that group to
organize themselves and achieve the goals that I set for them in a nonprocedural way, as
opposed to procedural. Procedural means I tell them how to do it. No. Nonprocedural
means you tell them what you want and let them figure out how to do it.That’s effective for many reasons. It allows the people you are dealing with to feel in
control, respected, to allow them to try things and be creative, and also extracts the
best out of them, and allows at the same time the esprit de corps that you create is one
of, gee, we’re participants, we’re stakeholders. We’re not supplicants or workers. We’re
managers as well. And that worked very well, and this group just responded beautifully
to that. There’s a price to pay from the management point of view. You now have a group
that you don’t control very closely, they have their opinions, and you’ve got to
basically adjudicate some differences among them, among their approach and yours. So
it’s a very fine line and a very interesting experience and a very gratifying one as
well, to work with such a creative, brilliant group and still extract the best from
them.
- FIDLER
- And how did you see your management philosophy relate to that of other ARPA PIs?
- KLEINROCK
- Okay. So the ARPA PIs were in touch with each other from their professional
relationships, but even in a slightly more formal way, we had principal investigator
meetings every so often, where all the PIs would get together—and there weren’t all that
many of us at the time. There were between fifteen and twenty of us that ARPA was
supporting. The interesting thing is that these other PIs were working in fields other
than your own expertise. For example, they were not all doing networking. They were
doing artificial intelligence; they were doing high-performance computing; they were
doing database; they were doing graphics, people I would not ordinarily have interacted
with. So I could observe and communicate and discuss and debate with them how they were
managing their projects, because they each had these large projects.I found a number of them had a similar philosophy. They have taken the same clue from
ARPA that the right way to do this—you know, the folks from MIT, from Berkeley, from
Utah, from Santa Barbara, they were all people who had enough confidence in themselves
that they could let other people do great work in their environment. They weren’t
challenged. In fact, they were enhanced by these other people they were working with in
their environment. So I found most of them having a similar philosophy, but remember
they were also influenced by their experience with ARPA, who was funding them and gave
them that freedom, and they saw how well it worked.But there was another group of talented people who did the same thing. As I told you, I
gave my own group, my own staff and programmers and developers, lots of flexibility.
Well, what did they do with that? Well, bless them, those graduate students formed their
own group of graduate students, well beyond the confines of the halls of UCLA. They
reached out to other graduate students at Utah, at UCSB, at Berkeley, at Illinois, at
MIT, at Stanford, and they formed their own graduate students’ mafia, if you will. They
formed this Network Working Group. They created—it came out of UCLA, but they all
participated in the Request for Comments group. They were active behind the scenes in
their own distributed control flexible environment, and so that same idea, that same
governing policy, beautifully permeated. They picked up the same message and they were
able to implement it as well.
- FIDLER
- Was there anything else from your personal, professional, or intellectual background
that influenced this management strategy? So far we have ARPA and we have your
experience on a more technical side with distributed network architecture. Is there
anything else that comes to mind?
- KLEINROCK
- Well, again, I spent many years at MIT and I was a graduate student, and I observed the
faculty there, and it was that same esprit de corps. There was confidence and, if you
will, respect for your peers, and between the faculty and the students it was a very
collaborative environment.It turns out at MIT that most of the gratification came from within your colleagues at
MIT that typically didn’t reach out across the world or across the continent, and so
there was a sense of these were Centers of Excellence here at MIT, and we’re here to
collaborate with, impress, cooperate with our fellow students and our fellow faculty. So
there was another sense that the way to work was not in an isolated domains, not in a
protective or secretive environment. It was a competitive environment for sure. All
throughout all the questions you asked me, yes, we were all competitive in a very
positive way, and we were willing to alter our direction if we felt somebody else had a
better idea rather than fight it and be secretive and pull it back. So, yes, the MIT
environment is another source of that.
- FIDLER
- And how do you feel that the operation of the Network Measurement Center compared to
other ARPANET nodes or even Centers of Excellence at that time?
- KLEINROCK
- Well, that’s hard to respond to because each of us had a different view as to what we
were doing with our ARPA funding. It turns out that the work here at UCLA was
specifically network-focused. Therefore, the network itself was the object of our
studies. In many of the other cases, like some of the artificial-intelligence work, the
work was not network-related. They used the network to collaborate, but they didn’t
study the network as an entity. So there was a difference in that sense, but the tool of
the network was the object of our study. Now, in terms of the management philosophy,
we’ve talked about that. I saw a similar distributed delegated approach.
- FIDLER
- Do you feel that constantly working with a distributed network architecture, as
something that not only you were working on, but the graduate students as well, that
that continued to inform the work and management philosophy at the Network Measurement
Center?
- KLEINROCK
- Yes, in the sense that it was self-feeding. It was self-generating. That philosophy took
hold and stuck, and it enlarged, as I say, by not only reaching out to the PIs, but to
the entire graduate student community. That group was an amazing thing to watch, because
in some sense it started what I might call, not to be pejorative, an underground
culture. These graduate students communicated with themselves in ways that did not pass
through the management of their respective universities.I remember one day going into the bullpen where many of the graduate students, mine and
others were working, and I saw they were very busy doing some damn thing or other. They
were on their terminals interacting with something. And when I questioned them, they
were all working on news groups, these hobby-related, interest-related groups that had
things to talk to each other about, be it recipes, be it photography, be it computing,
be it programming. And I was amazed. I was unaware that culture was vibrating right in
the bowels of this university here, but there they were doing it, and it was the
beginnings of social networking as we know it now. Certainly email, by that time, had
taken hold strongly. But it was an amazing thing to observe, and to suddenly be aware,
boy, this thing has reached out and has a life of its own.
- FIDLER
- And in the six years that the Network Measurement Center operated, that would have been
one change, the extent to which email was used, this underground that you described. Are
there other kinds of changes that you noticed between 1969 and 1975?
- KLEINROCK
- Yeah, a few things. First of all, initially, nobody wanted to participate in this
network. None of the PIs wanted to put their host machines on a network, for the same
reason that those three nodes at UCLA didn’t want to. They all felt their machines were
being used 100 percent of the time, they couldn’t possibly afford to give up any of
their cycles, and ARPA basically forced them to participate. “Since,” they said, “we’re
funding you, you’ll join this network.” We need you to do it.”As soon as they joined, they began to recognize the strength, but it didn’t happen—when I
say as soon as they joined, it didn’t happen immediately. The early usage of the network
was spotty, and the reason is that it was very difficult to use any of the external
resources through the network. I mean, what do you want to use? You want to use
hardware, software, applications, and services. Well, in order to do that, you have to
log on to a remote machine, a machine with which you may not be familiar, so you’ve got
to get a logon name, you’ve got to learn the command language, you got to learn the
services, you have to learn the way to use that remote facility. Why bother? Why not use
the machine here in your local environment that you’re used to? Well, if they have
really excellent resources, you’re going to push, but there was no easy protocol by
which you could access these remote machines.So in the early days, most of the usage of the network was by people who had left one
locale, one university, and taken a job elsewhere and wanted to use the machine at the
original site they came from. They had a logon, they knew the command language. They
knew that better than the new place they had arrived at. So there was that kind of
reaching back to your old environment that caused some of the early traffic.The Network Measurement Center also generated a lot of traffic, which was not true
traffic; it was basically measurement traffic. It was very hard to use. So as it became
easier to use the network, and what made it easier was once we introduced the host-host
protocol in 1971, that we began to see sites accepting the use of the network as
something powerful and people actually beginning to do something useful. So those
changes took place in those early seventies when it became easier to use, and we began
to get more interesting sites attached.By the summer of ’70, we had already crossed the United States in the network, but we
needed some more interesting sites to develop, and by the time you get some of these
large machines attached, some of what we now call super computers, the big mainframes,
with real resources that you really want to use, that happened through the early
seventies. Those first four sites, there were some specialized resources, but they
weren’t really by themselves something with enough variety and capability, that you
really want to learn networking. You learn networking with a lot of stuff out there, and
that happened in the seventies as well.
- FIDLER
- Do you remember the first time that you began sending electronic mail over the network?
- KLEINROCK
- No, I can’t. I can’t tell you what the first email message is that I sent, but it was in
late 1972 once Ray Tomlinson had introduced it. The amazing thing about what Ray did is,
time-sharing systems had email for years. Email was nothing new, but it was locked to a
particular time-sharing system, so only the users of that time-sharing system could
exchange email. What Ray did, he said, “Let’s put that on a network,” so people at
different time-sharing systems across the network could share email. The interesting
thing is the way he announced the arrival of this email capability was through the first
email he sent to his colleagues at BBN. He said, “There is this capability now, network
email,” and it was the first email, which I think is beautiful.
- FIDLER
- Just to go back and finish up with our discussion of management and management
philosophy, is there anything that you want to add about the way that goals were set or
tasks were defined, either by you or within the groups that you managed?
- KLEINROCK
- Okay. So there were some large goals that were clearly set early on. We needed this
host-host protocol; we needed to get the host-IMP protocol going; we had to connect to
the IMP when it arrived; we had to get Network Measurement Center running experiments.
Those kinds of high-level goals I laid down as things we needed to do. But as I said
earlier, the way we had those implemented was to have my group—the software development
group was headed by Steve Crocker, the hardware group by Mike Wingfield, I headed the
research group with my own Ph.D. students. We let them basically direct their own
activities, largely.I remember Steve would come to me every so often with two of his strong cohorts, Vint
Cerf and Jon Postel, and they’d present to me a travel budget they’d want, and it was
always far more excessive than I felt they should be asking for, but more often than
not, I granted it because I knew they had to travel to the other sites as they were
gaining access to colleagues at their level, the network working group. So we have a
discussion and we talk about what they were doing, but the details of the software that
they were developing and how it was being developed, I participated—I let them do it. I
didn’t direct them at all. They were better at software than I was. I was better at
performance evaluation, networking, etc. So I gave them a great deal of flexibility, and
they took it on very well.The research group I managed very closely. They were my Ph.D. students. The measurements
I conducted very carefully. We set up the experiments. Some of my own research Ph.D.
students were doing a lot of the measurements themselves.
- FIDLER
- Can you say more about the different groups? You’ve touched briefly on research software
and hardware, but the way that those groups interacted at the Network Measurement
Center, the different kinds of things that they did, the way that they worked.
- KLEINROCK
- Well, in addition to the protocol aspects of it, we had people doing some programming to
create the software for the measurements, so we had a programming staff as well. So the
groups I identified, there was software development for the protocols; there was
programming to support the Network Measurement Center activities; there was a hardware
group and that was a very small but very important group to get us connected to the IMP
in the first place, maintain it, and maintain the facility we had in terms of our center
here. Then there was the Ph.D. students I supervised in networking research and a few
related faculty who were participating and interested in networking at the time.I remember holding a class on some of the theoretical aspects, and in the class were not
only my graduate students, there were some faculty attending to get up to speed on the
networking technology that was being developed in those early days.
- FIDLER
- And did the number of these groups or their content or their focus shift over time
between, for example, the establishment of the Network Measurement Center and then its
discontinuation in 1975?
- KLEINROCK
- Well, understand these people that I’ve mentioned, Steve Crocker, Vint Cert, Jon Postel,
Charley Kline, Bill Naylor, they were all Ph.D. students and they had their own Ph.D.
research to conduct. Some of it was related to and coincident with what they were doing
for their dissertations, some was not, or what they were doing on the Network
Measurement Center. So to the extent that it was not directly related to NMC stuff, they
began to work more closely with their supervisors, less with the urgency of getting some
of the critical protocols and connectivity up and running early on in our work at the
Network Measurement Center. So the group continued to operate as an entity, but they were each going in individual
directions as they were pursuing their own Ph.D. research, and not all of those in the
group were Ph.D.’s of mine. They were Ph.D.’s of other faculty members as well, quite a
number. So there was a sort of separation of some activity, but the group pretty much
held together.We were busy running the experiments, interacting with new nodes. The Network Working
Group took on a life of its own. They were looking at things that were extending well
beyond the Network Measurement Center’s responsibilities, so that was not directed out
of my domain; it was directed out of that Network Working Group domain. So they began to
function almost as an entity in that domain on their own.
- FIDLER
- In retrospect, it’s easy to look at the work that was done and accomplished as
inevitable progress and development. Can you speak to the perception of risk of this
research and even how it might have changed over time, for example, the viability of
getting this network running in 1968 versus the kind of work you were doing in 1975?
- KLEINROCK
- Sure. In those early days, this concept of making a computer network, a packet-switched
network, was a challenge, a serious challenge from an engineering point of view, from an
implementation point of view, from an up-and-running acceptance point of view. In my
mind, there was no question but that the technology would work, but would it be properly
funded, would it be properly managed, would it be properly implemented, would it be
properly supervised, were challenging questions. And all of us involved in those days
took it on as a serious challenge and an engaging and exciting challenge.We recognized risk in the sense that, yes, we had to do our job well to make this come
off properly, but we didn’t get a sense that we were in danger of failing. It was a
question of how well would it work, how exciting would the challenge be, would it gain
traction, that was always a question. The original concept of this network, we
originally designed the 19-node network. That was in a spec that went out with the RFQ.
And BBN, who won the contract, originally made it possible for 64 nodes only. They had 6
bits of address space.So the original implementation mentality was not one of a very broad-based network. In
terms of the vision that we had, yes, we saw this growing, but the implementation
direction didn’t head that way directly. So the risk of failure was small, in my mind,
and, I believe, in most of the other participants, the risk of failure of the
technology. But failure of the impact of this on the digital community, that was always
a question. Would anybody use it? We’ve worked very hard to make it easy for people to
join. I told you originally many of the PIs did not want to join, so we, as we developed
the protocols, we removed as many of the impediments to join in this network as
possible, to make the protocols as easy as possible, to make the requirements for
membership as simple as possible, to make the restrictions on what could be run and what
could be done on the network as simple as possible.That’s one of the reasons that we failed to put in any protection against adverse
behavior. We knew everybody that was coming on to the network. We trusted everybody.
Those early days, we had an ARPANET directory of all the email addresses, when email
came in. We knew everybody who was on the network. So, as I say, we trusted them. That
led to the lack or zero focus on security and protection against what later became the
dark side of the network, but we wanted to make it very easy and collaborative, and so
the idea of trust, openness, sharing, participation was dominant in the culture, and
that is what allowed people to come in and join and make it easy for them. And even
then, it was not a rapid uptake, because until the host-host protocol came in, it was
very difficult to do inter-host communications.Actually, there was a rather interesting point. Bob Kahn took over as director of IPTO,
and he was also responsible for creating a public demonstration of the ARPANET in 1972,
in October 1972. I mention that because up until that point, many of the sites were not
aware of what the others were doing. We didn’t really know what other applications were
in the bowels of the other universities. They were doing their own applications, really
neat stuff. The challenge for this 1972 demonstration was to bring all these sites
together in a public arena so that the public, meaning the other engineers and other
researchers, could come and see how to use this network. So we all collaborated on
putting our applications out there in ways that other people could access them through
the network, and we made joint efforts with other groups to make collaborative
applications as well.So we began to learn what was being done out there. The artificial-intelligence community
had a lot of interesting applications, chess-playing programs, robots that could run
around. The simulation world got together. We were doing a number of things. One of the
early demonstrations was to run a distributed simulation of air-traffic control across
many computers in a dynamic fashion. I won’t go into the details now, but it allowed we
PIs to understand far better what was going on on the network in terms of applications
and allowed the public to see it as well.So at that point, there was a greater uptake and appreciation for what was available to
us as the participants in the network and, if you will, the designers and the
contributors of applications and services to the network that we never had before. So
October ’72 was a rather important and successful demonstration.
- FIDLER
- So in terms of growth and change and patterns of use of the ARPANET, October 1972 is one
major shift that you saw. Were there others over this period of the Network Measurement
Center where use changed significantly?
- KLEINROCK
- Well, you have to understand that the growth of the number of hosts on the
Internet—eventually called the Internet—was exponential from day one. So when you ask
for significant changes, there were no major inflection points for traffic, but there
was this kind of S-curve behavior. When you get a growth of a thing which makes the
network more interesting to people, that becomes an S-curve, and another inflection
point from another technology, that collection gives the exponential behavior. So, yes,
there was some things. As you say, the International Conference on Computer
Communications, 1972, that demonstration was an important one. The host-host protocol
was another one that made it possible whereby people could now gain access. The bringing
on of some other networks like the radio networks, the ALOHA Network, the Satellite
Network, access to Europe via the link to Norway and London, they were all important in
the sense that it continued to reinforce the fact that this network was a growing
network full of good things of interest, it had staying power, and the community
continued to enlarge in participants, and the participants were all the researchers in
the field of digital computers. That community just grew and grew and grew, and by its
own growth, added to the power and the steam of this locomotive called the ARPANET,
hence Internet.
- FIDLER
- You mentioned earlier trust and openness and sharing as being integral to both the
working groups at UCLA, the people working at the Network Measurement Center at UCLA,
but then also you suggested that that was being built into, I guess, patterns of
behavior on the ARPANET itself, at least in the early days. Is that accurate?
- KLEINROCK
- Yeah, it is true. Across the ARPANET there was a sense of camaraderie, there was an
implied, implicit netiquette that developed. It wasn’t ever written down as a set of
rules. It was an accepted mode of behavior, and people behaved well. People didn’t—what
should I say—blaspheme. Certainly spam was not introduced.We recognized this was an experimental research network among colleagues that we
respected, where this was an engineering effort. This was not a commercial effort, this
was not a financial bonanza, this was not looking to exploit; this was looking to
create. And it was a golden era of creativity. You couldn’t have asked for a better
environment. It didn’t have the trappings, as I say, of a business side, a heavy control
side, a nasty competitive side, some freaky people doing bad things.It’s amazing, now that you ask this question, that I look back on it and that all of
those bad elements did not participate, did not manifest themselves in those early days.
It was a group of really well-meaning, well-minded, well-behaved people all doing
amazing work. I think the fact that we were doing good work and getting the
gratification of creating and growing and seeing this thing take off kept things on
course. That was enough, sort of was the container.
- FIDLER
- So you’re saying that having this ability to directly work on and contribute to this
growing community is something that you saw keeping behavior in check.
- KLEINROCK
- I believe so, but it wasn’t clear to me until you asked the question, somehow. It’s sort
of like a tornado that stays collimated because it feeds on itself. And there was enough
gratification, enough success, enough creativity, that we didn’t need other sources of
diversion or whatever to make it exciting. It was exciting.
- FIDLER
- Now, you mentioned netiquette as an unwritten set of rules for, I suppose, conduct and
behavior. Did you see a similar level of, like, unwritten practices at UCLA at this
time? Was there a similar extent to which ways of doing things were developed that
weren’t necessarily codified, but were, nonetheless, prevalent?
- KLEINROCK
- Well, that harkens back to what we talked about earlier, the idea that it was not a
heavy-handed management. It was a cooperative, it was a respectful, and a creative
environment. Much of that was manifested in the research groups that we had, but not so
much across the entire department. The entire department was not participating in
network effort. These towers of excellence in the department at the time, one of which
was this whole networking world, there were others as well, and it didn’t
cross-fertilize as much. So in some sense, this environment that we’re describing was
not even across the entire department, but it was across the entire community of
like-minded people doing work.
- FIDLER
- You mentioned traction when we were talking about risk, not will this work technically,
but will this be successful in certain communities. Was there a moment when you said,
“Okay, this is definitely going to take off,” more than just saying, “This is what we
hope it will be,” but something where you knew, “Okay, nothing can stop this now”?
- KLEINROCK
- Well, I can’t tell you when I first got that sense, but certainly by the time NSF
entered the picture, and NSF entered the picture in an interesting way. They created
these super computer centers in the early eighties that were not necessarily connected
to each other, and we were aware of the super computer centers. They were not an
important part of the Internet at the time, but when NSF decided to connect them
together using the Internet, suddenly the constituency of the membership in this
Internet community increased dramatically from the computer scientist closed group that
we talked about earlier, namely to the group who was being supported by ARPA, to a much
larger community, a community of chemists, physicists, physiologists, archeologists,
oceanographers, etc., all science.At the same time, it was around that period when those groups that were not being funded
by ARPA, those computer science departments, for example, that were outside this chosen
few, if you will, decided they wanted to participate too. So some efforts were started
which were successful to allow them to participate in the Internet experiment, namely
CSNET. Computer Science Net was formed, PhoneNet was formed, other ways to gain access
without being blessed by the large ARPA-funding machine. So they began to come in on
their own. They began to extend beyond the ARPA-selected community, and I think that was
an important point when I recognized this is now growing on its own. It doesn’t need the
constant stoking of ARPA funding. It’s reaching out. So that certainly was a very
important development at the time.But I must tell you that email, which was 1972, was a strong impetus to getting other
people excited about this, because that drew in people. They didn’t know about
networking, but they sure liked this email thing, and they were using email without even
necessarily knowing they were using a network, and so the community enlarged that way as
well. But until we get the breadth of the NSF world, the community was largely computer
science efforts.So when did I first realize this thing was unstoppable? The first thing when I mentioned
earlier is when we went international and we went more than just wired networks. We went
to ALOHAnet. We went to Satellite networks. The whole packet radio effort, by the way,
interestingly enough, the idea of ground radio packet-switching, was started in the
early seventies. This is very early in the genealogy we’re talking about here. The
applications there were broad-based. They were certainly military, but certainly in the
civilian world as well, what with communications coming into fashion. That whole effort
began to have a life of its own, but it was coupled into the ARPANET because it was
using the ARPANET for communications as well.I could see then that this was not just about wire line machine-to-machine
communications. It was allowing other things that we didn’t anticipate to suddenly
engage in this thing we call networking in a way which would cross continents and not
just hundreds of feet or hundreds of miles. So you could get a sense that it was gaining
access, the international side and the broad-based other network side, and suddenly the
international community, even though they didn’t engage heavily in terms of
participation, because across Europe there were a number of networks that started to
spring up. The Nordic Network, for example, which is a network based on high-voltage
electric utilities, wanted to gain access. The French created their CYCLADES network,
and CNET was the research effort in France, and the British networks and the Spanish
networks and the Italian, they were beginning to emerge and then attaching as the
Internet grew. So you could see this had a life of its own at that point. This is
through the seventies and the eighties. So I can’t say exactly when, but every indicator
was up that this is going to grow.
- FIDLER
- Certainly. Another element of traction is program transfer, is getting the ARPANET away
from ARPA at a certain point. In 1975 it went to the Defense Communications Agency. When
did you see transferring the ARPANET to a different operator as being something that
should happen or that people wanted to happen?
- KLEINROCK
- Well, it did happen, as you say. The Defense Communications Agency, DCA, took over the
Network Measurement Center in 1975. DARPA was funding the whole boat for a long time,
and it was clear they were eventually going to step out of that role. So other groups
began to think about taking over some of the management function. Did I think it was a
good thing? I’m not sure I had an opinion in those early days. I was sorry to see the
Network Measurement Center transfer to DCA, not so much from a personal point of view,
but because I could see that they were not picking up the role. They were not conducting
the kinds of critical experiments we were doing. They didn’t have the mindset to ask the
right kinds of questions. They were doing network monitoring, at best, and then they
stopped doing that, instead of asking questions, well, what would happen if we stressed
the network in the following way, and what do we expect will happen before we conduct
that experiment. So that whole approach to scientific experimentation disappeared. For
that reason, I was regretful at that point.In terms of what ARPA’s role was at the high levels within IPTO, within ARPA itself, once
you get out of IPTO into the ARPA level, what’s now called DARPA, that was beyond the
world that I lived in, and that was managed by administrators and funders and
politicians and government agencies, which was not a world I was involved with and not a
world that I cared that much about who the players were as long as it continued.And so, yes, the PSIs appeared, the other network operators appeared. I forget exactly
when one of the access networks appeared. We had one of our own here. What was it
called? What’s it called? Los Nettos. I can’t remember the names now of the early access
providers, but it was a group trying to provide local access at high speeds, and it was
local, and suddenly those things became commercial. First, it was a group of researchers
putting it together. Companies suddenly went out there and became a commercial entity.
You could see at that point that the financial world was now taking note of this thing
and trying to make a business out of it, and certainly they did, and it was in the late
eighties when these things began to take on that commercial side.
- FIDLER
- We speak alternately about, on the one hand, the Network Measurement Center at UCLA, and
then on the other hand, the UCLA ARPANET node. Maybe you can speak to more about how
those two terms overlap and how they’re different.
- KLEINROCK
- Well, the UCLA node, the ARPANET node, was the first node on the ARPANET. Its role,
besides being first node, was to run experiments, and being the first node, we were able
to begin experiments from the first node on. We continued to serve as the Network
Measurement Center. The fact that we were the first node was almost irrelevant in terms
of there was a second node, a third, a fourth, etc., and number of the node didn’t
matter. The function of the node mattered. Ours was Network Measurement Center, SRI was
a database capability, Utah was graphics, etc.So the point is a node was basically an IMP connected to a host. The IMP had a number:
one, two, three, four. We were number one. The host provided the services and the
capabilities and the applications. We were the Network Measurement Center with a host.
Utah was a graphics center. So the most stark separation is that the IMP was the node
and the host was the service, and yet the group of people working there, like at UCLA it
was that group we talked about, those forty-odd people that I had put together to manage
the network.
- FIDLER
- You spoke about measurement previously. Maybe we can speak about the kinds of network
measurement experiments that you ran.
- KLEINROCK
- Sure. So the plan was to create a 4-node network in the first four months. September,
UCLA, September ’69; October ’69, SRI; November ’69, UC Santa Barbara; December ’69,
University of Utah, a 3-node network, a little triangular network and a stub off to
Utah, and then stop for about three months while we at the Network Measurement Center
and those at BBN tested out this test network. BBN was measuring the quality of the
lines, some of the traffic, were the IMPs up or down. At UCLA, we were trying to send
flows, traffic through the network to see the response time, the throughput, the buffer
utilization, etc.So we conducted some early experiments. The most obvious first experiment was to pump
data from the first node to the second node, namely from UCLA to SRI, to see how much we
could pump through and how long it would take to get files through. Recognized this was
in the 4-node network or whatever. And so we ran some very early throughput experiments
from UCLA to SRI by sending just one connection from UCLA to SRI and then increasing the
number of connections to see how much throughput we could achieve. Those early
experiments were very informative in terms of phenomena that we had anticipated, but
weren’t sure how they’d behave in terms of the throughput themselves. So we were happy
to see that we could get good response times and good throughputs through that early
network.I’m looking right here at a page from my book where I show the throughput between UCLA
and—oh, it’s UCSB, I’m sorry, not SRI. My mistake. It was from UCLA to USCB, from the
first node to the third node as a function of how many connections there were, if you
will, generators between UCLA and UCSB. And no surprise that we could begin to approach
up to 50 kilobits per second as long as we had single-packet messages, but as we
increased the number of generators, we began to go above 50 kilobits per second, which
was the line speed connecting UCLA, UCSB, because we, in fact, forced some parallel
routing through SRI, which was an alternate path.So we studied this alternate. It was very smooth behavior, and we were able to get up to
about what we expected, two paths from UCLA to UCSB, subtract the overhead, and you get
about what we predicted. So that was a very nice experiment, and we got reasonable
response times as well.The network was functioning, there were no major faults, and we began to run some larger
experiments as we connected more nodes into the network, and then we began to stress the
network. We began to do a variety of things which caused the network to fail, and I can
talk about some of those experiments if you like. For example, one of the things that we
noticed early on, something that was predictable ahead of time by people like Bob Kahn
and ourselves here, was something called store and forward lockup. It turns out that if
two opposing traffic streams were heading toward each other through the network, they
could occupy buffers in a certain way which would prevent either one from getting
through unless one of them relaxed, and neither one would relax. In that case, we got a
lockup or a deadlock. It was called store and forward lockup.We also found lockups in some of our other experiments—and once we decided that, we
determined how to fix it. So we would notify BBN, “We have this lockup. Please fix it.”
It would take them six months to fix it because they were busy keeping the network up
and running in terms of monitoring the lines and the IMPS.We ran a number of measurement experiments, whereby UCLA would send out a number of tests
to various nodes, and take snapshots of what every node looked like, and send that data
back to UCLA. Now, taking those snapshots was easy. Sending all that data back to UCLA
all at the same time clearly placed a stress on our network. All the traffic was coming
back to one node, UCLA, and, sure enough, we crashed the network. So we were conducting
this innocent experiment, taking snapshots, and down went the network. And, of course,
BBN noticed that because they were monitoring the status of the network, and they called
us up and said, “UCLA, you just crashed the network.”And we said, “All we’re doing is taking measurements.” So BBN brought it up again, we
continued the experiment, down it went again. So we decided we’d better find out what’s
happening, and we found out it was due to a kind of allocation of buffers here at the
reassembling site that happened to be a kind of error in the protocol. We told BBN what
we suspected was the case, and they fixed it.And as we conducted other experiments—we had christmas lockup, we had piggyback lockup,
etc., a variety of different types—we began to get anxious that we’d like to know what
was the operating system that the IMP was running, what was the software protocols they
were running, so that we could look at the code and decide how to fix it. And BBN said,
“No, you’re not allowed to have that code. It’s our proprietary code.”And we said, “What?” [laughs] And they held on to it.So we pointed out to ARPA that this code ought to be open. ARPA had paid for it and we
needed it to conduct our measurements properly. So, finally, ARPA leaned on them
heavily, and BBN relented.So the next time we found some errors, we’d run an experiment, we’d stress the network,
intending to break it—we could break it—then we’d find out why it broke, we’d show BBN
how to fix it because we had the code. It would still take them six months to fix
it.So as you can imagine, there was a kind of tension between the group at BBN, and the
group at UCLA. We were breaking a network, we told them how to fix it. It was sort of,
in some sense, that they considered it to be their network because they had deployed it.
So how quick will they fix it? We were pressuring them, and they were complaining to us.
So the relationship was one of a little bit of tension, but differing motives. They
wanted to keep the network up. We wanted to take it down.
- FIDLER
- Right. You saw it as an experiment in progress. They wanted to have a functioning
network.
- KLEINROCK
- Exactly. Exactly right. For them it was not a business but an engineering role, but so
it was a kind of—what should I say—professional competition. It was not mean-spirited.
It was almost fun in a way. We—what should I say—annoyed each other in ways that
produced progress. But in terms of the Network Measurement Center, that was our role,
and I was unhappy when it couldn’t continue.
- FIDLER
- In 1974 you published a paper with Bill Naylor, “On Measured Behavior of the ARPA
Network,” and there, amongst other things, you found, I guess, favorite node pairing
would be one way to put it, and also something that you coined “incest.” I wonder if you
could speak to both of those properties.
- KLEINROCK
- Sure. So we conducted rather extensive measurements of the traffic on the Internet. Bill
Naylor was my Ph.D. student. His dissertation was all about network measurement. So we
made these measurements to see what was going on in the network, and we wanted to see
what we could extract out of it. Just as people do data mining today, those were the
early, probably one of the first data-mining experiments ever run.So we noticed some things in the statistics of the way the traffic would move around the
network. For example, we discovered that a significant percentage of the traffic at a
given IMP was going between two hosts at that same IMP, namely host at a given node, say
UCLA, was sending traffic nowhere in the network. It was going into the IMP and back out
to another host at same location. After all, if two machines want to talk to each other,
what better implementation than to use the local IMP, which knows how to connect two
machines together without even going out of network. I forget the number. I think it was
21 percent of the network. You can remind me, Brad.
- FIDLER
- I think there was an average of around 20 percent of zero hop.
- KLEINROCK
- We were very much surprised. I’m checking in my own book right now, what the number was.
Yeah, basically a lot of traffic was going zero hops, and I called it incest because it
was between two siblings at a given site. And, yeah, approximately 22 percent of the
traffic was incestuous, a huge amount, one-fifth of the traffic was incestuous. And we
found, in addition, that certain pairs of sites were very heavily intercommunicating,
and we noticed that as well. So there were some favorites. For example, UCLA would have
a favorite—I forget who it was—and SRI would have a favorite. Yeah, 34 percent of MIT
traffic was incestuous. As an example, the favorite sites—I could look up again, not so
easily, but various nodes had various favorite sites. So we began to extract some rather
interesting behavior through the network. In fact, we realized that if we had links only
between a given node’s favorite site and no other node, that that would account for a
significant fraction of the traffic reserved on the network, which is an interesting
observation. You don’t do much with that. You want the full network topology so that you
can communicate with everybody.
- FIDLER
- And there was surprise when these results came in, because, on the one hand, you’ve gone
to all this trouble of making an invisible subnet, and it should be the same cost to
communicate to any other node, but then you have these patterns of local use.
- KLEINROCK
- Yes. Well, in hindsight, of course, it’s not a surprise. There were certain sites that
offered better services than other, or there were certain pairs of sites that had a lot
in common. Just because a node is in a network and easily accessible doesn’t mean it’s
one you want to connect to or one you want to glean services from. So, in hindsight, it
makes a lot of sense. Some of the larger sites were more interesting. Some of those that
had more interesting applications, be it artificial-intelligence programs, simulation
programs, were accessed quite a bit more. But the incestuous traffic was a surprise. In
hindsight, it makes sense, yes, you want two computers at a site that know each other to
communicate. Fine, use the IMP. But we did not anticipate that. We thought a node was a
node, and they wanted to talk to other nodes, didn’t want to be incestuous, and that was
rather interesting and kind of a remarkable—one-fifth of the traffic. That was a lot.
- FIDLER
- So that you told you about the way that this network was socially organized, I suppose.
- KLEINROCK
- Excellent observation. You’re exactly right. If I knew somebody, I want to communicate
with their machines that they were producing applications on. And as the network grew
and we began to interact with people remote from our physical location, as distance
disappeared, that effect goes away. I wonder what the incestuous traffic is now. On the
other hand, we have so many hundreds of millions of hosts. I suspect the incestuous
traffic is tiny. I mean, who wants to connect to my laptop except me?
- FIDLER
- And Ethernet came in as something of a solution to this local traffic.
- KLEINROCK
- Yes. It’s interesting that it was recognized, not only from our recognition of
incestuous traffic, but as the digital community matured, that more and more machines
were being used at local sites, say at corporations, say at universities, say in
government offices, and there was a need to connect things based on locality as well as
on what services they have to offer. It would be natural that people want to communicate
within a given organization or department.And so when Ethernet came in, which allowed for local area networks to emerge in a
dramatic way, that sort of provided yet more motivation and more access to incestuous
traffic at a given site, where the Ethernet link was the communications among siblings
at a given location.By the way, interestingly as well, perhaps one of the first really financially successes
that took place on the Internet as a communications entity was Ethernet itself. I would
say that Bob Metcalfe was one of the earliest wealth-accumulating individuals on the
network, way before many of the other networking successes came along. I mean, IBM and
all those other companies are doing great in the digital world, but not in the
networking world.And this whole local area network is also interesting because for years the trade
magazines were predicting this is the year of the LAN, and it kept being postponed and
kept being postponed. And I’ll never forget, I think it was 1982, but I’ve got to check,
when finally the IEEE, who, was tasked with the job of creating a local area network
standard, came out with their standard, and I remember very clearly mentioning this in
many of the public conferences and trade-show arenas. I would produce a foil, an
overhead foil, of a page in, I think it was Computer World, and the headline was “IEEE
Adopts Local Area Network Standards,” and the most important letter in that sentence was
the last S. They didn’t adopt one standard; they adopted three. They adopted carrier
sense, multiple access on a wire named Ethernet, with collision detect. They adopted
Token on a ring, which was the IBM, SNA, Token Ring technology. And by the way, Ethernet
was on a bus as opposed to a ring. And just to be safe, IEEE adopted Token on a bus. So
they had all the combinations there, but it was hilarious that they adopted all
three.And for a long time, the battle was raging between Token Ring and Ethernet, and one came
out of an industry proprietary standard, Token Ring, IBM’s, and the other was a
grassroots effort, this Ethernet bubbling up from beneath, and we all know who won. It
was a grassroots effort that made it, and that’s a story that’s repeated over and over
again. An industry-forced standard from above will almost always lose to a grassroots
effort from the user population from below. But you’re quite right that Ethernet
exploited and recognized and made it possible for this location-based communications to
blossom. [End of July 20, 2013 interview]