Larry Roberts | A pioneer looks to balance Internet scale

Along with Vinton
Cerf, Robert Kahn
and Leonard Kleinrock,
Larry Roberts
was one of the
founders of ARPAnet,
the network developed
by the Defense
Advanced Research
Projects Agency that
later became the

Forty years after
Roberts began working
on ARPAnet, the
Internet is being
overloaded with
video and peer-topeer
which it wasn't designed
for, he said.
Roberts also is the
founder and chairman
of Anagran, a
company that offers
products to manage
IP traffic.

GCN:Did you have any clue
the Internet was going to
change so many facets of
modern life?

ROBERTS: I had this perception
that we would make
knowledge available instantly
to everybody in the world. It
was getting to be all on computers,
and computers were
incompatible. And I thought,
we need to move from language
to printing press to
computer network. Now, the
Web is doing that, but we're
doing a lot more. The personal
communications, the
video, the other things, we'll
get to, but basically, that wasn't
the first thought. Before I
did the network, there was
the knowledge network.

GCN: What do you mean by
a knowledge network?

ROBERTS: All the computers
were incompatible.

We had no way to move stuff.
Even at [the Massachusetts
Institute of Technology] I
had various computers at
various places, and to move
stuff between them, we had
to find a compatible media, a
tape that worked in all of
them, which was not always
easy. So I started doing
experiments there, proved it
worked, and we did it in a
network after laying out the

GCN: You were sending the
first e-mail messages, right?

ROBERTS: Ray Tomlinson
actually built the file transfer
mechanism called Send
Message and Read Message.
He started sending me stuff
because I was in charge of
the program. And it was like
a teletype that was pouring
out. It wasn't very functional,
because you couldn't reply,
you couldn't look at the message
names, you couldn't
store, you couldn't forward.
So I read the first e-mail
envelope in 1971. E-mail
looks about the same today.

GCN: Those were pretty
big computers.

ROBERTS: I actually had
a DEC minicomputer in my
office that was the size of a
filing cabinet. Most of the
people on the network had
bigger computers. Everybody
had different ones.

GCN: Could you explain the
concept of Flow Manager, a
load-balancing device from

When I left
the Information Processing
Techniques Office at DARPA,
they finished TCP as a protocol.
They got that on the network
in 1983. Then by 1986,
we realized it needed to slow
down when things got lost
because the network threw
things away if it was overloaded.

From 1986 on, it was
stable, because TCP would
balance with the switching
equipment, which was just
dumping random packets.
Now dumping random packets
hurts the voice, hurts the
video and makes it almost
unworkable. It doesn't work
with them, because they're
fixed rate, so they won't slow
down. So it doesn't even help
to discard the packets. And
TCP will slow down, but it
almost comes to a grinding
halt and stalls. It's a very
awkward system in response
time, the way it's working.
What I'm trying to do with
Flow Manager is make a
change in the way the networks
undo the controls. In
other words, it's got to balance
with TCP.

The latest thing we're seeing
is that peer-to-peer is
overcrowding the network.
Basically, P2P is unfair, in the
sense that if you have more
than one flow, you get N times
the bandwidth, and every
flow gets equal bandwidth, so
you keep adding flows and
you do better than everybody
else. That's unfair to everybody
else who didn't use it.
And so we've come to the
conclusion that the average
rate ... should be the same for each home or for each user. It
doesn't matter how many flows
you use. So P2P can do its
thing, everybody can do their
thing, but they get fair use of
the network. We don't look inside
the packets to do that, we
don't go searching and violating
privacy, we just control the
rates of the flow. So that turns
out to be a radical improvement.
It makes it fair, and it almost
stops the net neutrality fight,
because now you don't have to
argue about it.

GCN: How did ARPAnet

We put four
nodes in, starting in October
1969. By the end of the year,
we were on the West Coast:
the UCLA, Stanford Research
Institute, the University of
California at Santa Barbara
and the University of Utah.
UCLA was the first, SRI was
the second, then we quickly
expanded it nationwide in
1970. By 1971, we did a big
demo, which illustrated to all
of the press and the public
that it worked, and they were
shocked by it. The research
people had seen it, but the
world hadn't really. And the
telephone people who had
been saying it would never
work finally realized that it
would. And they threw rotten
apples at us. They couldn't
believe that this would work.

GCN: So it was all text?

It was text and
data. We transferred binary
data for photographs. We did
almost immediately start testing
voice. And voice had low
enough bandwidth that there
wasn't a big problem, we knew
that would work. But we didn't
think video would ever be feasible.
We knew it would have
to be vastly faster, like 100
times faster, like it is now. ...
Nobody saw video coming.

GCN: It wasn't designed to
carry video?

It's still not
designed for video. It doesn't
scale. Changes will have to
happen to make video work
over the network in the scale
we're using it. That's why we
have to potentially block other
users, if there are 10,000 other
people using it.

GCN: What do you think the
Internet will look like in five

In five years, I
certainly hope we have a significant
deployment of systems
that protect us and provide
equality, as I said about peer to

Secondly, I would hope that
we are managing video properly,
so the kind of Flow Manager
things we're doing will help.
They will get deployed over five
years, I think, by somebody.
Thirdly, I don't know how
we're going to achieve it, but
we have to add network security.
I want the network to tell
me that I know that that's you
I'm communicating with and
you've been verified. So you
would have to have then a certified
ID when you got on the
network. And then everything
you did would be approved,
and I'd know who you were
and I wouldn't get spammed.

About the Author

Trudy Walsh is a senior writer for GCN.


  • Records management: Look beyond the NARA mandates

    Pandemic tests electronic records management

    Between the rush enable more virtual collaboration, stalled digitization of archived records and managing records that reside in datasets, records management executives are sorting through new challenges.

  • boy learning at home (Travelpixs/

    Tucson’s community wireless bridges the digital divide

    The city built cell sites at government-owned facilities such as fire departments and libraries that were already connected to Tucson’s existing fiber backbone.

Stay Connected