So, welcome back. Now the purpose, of
course, of all this stuff that you had in
your office, and all that banging, and all
the sqweeee squealing noise where we're
sending the data across the across the
phone using sound. All the purpose was the
fact that computation was rare and
extremely valuable. And for scientists who
were trying to solve research problems,
access to computers was essential. And you
couldn't all sit, sort of in a little ring
around the computer. It's so we would have
phones in our offices and we would, you
know, work in our office, just like I'm
working in my office right now. But there
wasn't enough computation [inaudible], any
work in my office so I had to connect to
something outside. And so this was the way
of science. And, it, it, it was the fact
that comp-, computing was rare, and access
to computing was a critical enabler of
scientific research. Now, I'd mentioned
earlier, data transfer with leased lines.
And so, while it. You tended to interact
with the computer that was rather local,
geographically to you because you could
have this permanent dial up connection all
day long without paying a permanent
charge. If you were a bank, or you had
some really critical need you would lease
some line from the phone company 24 hours
a day, seven days a week so you could send
data across that anytime that you wanted.
No dialing, it's always connected and,
after while, I mean you can send data, we
academics wanted to communicate with each
other. It would be nice to be able to use
each other's computers, but we tended to
have too much, but sometimes we don't want
to send a file, or some email, or
something like that. And so this led to
the invention and the creation of store
and forward networking. And how this would
work is we would sort of, you would sort
of have some thing and you would use a
modem to do all your dialing, like that. I
mean maybe they had paper and, and didn't.
That looks a little too advanced, but you
would have some geographically local
computer that was your, sort of the compu
ter that did most of your work. You didn't
have a computer in your office, you just
had a connection to the single campus
computer or on a few campus computers. And
then what universities would do is they
would lease a line. And then we could
send, write a mail program that would run
on this computer and then would send mail
and then everyone else would read it. And
what happened was is we sort of started
stringing them together in these snakelike
structures and so, we could share this.
And so let me just show you kind of how
the store and forward networking works. So
somebody sends a mail message in. Now
let's say, let's say we are this bottom
person down here, okay. And so someone
else has sent a mail message in, they're
sitting in there. And now the next person
sends a mail message in and now ten
seconds later, you send a mail message in.
Those mail messengers are sitting in a
cue. They're waiting just like waiting in
line at the bus stop or waiting in line in
a train station. Waiting in line for a cup
of coffee at Starbucks. They're waiting in
line, and what would happen is the
computer that was our local computer would
then start sending that data across the
line. Okay, and slowly but surely it would
take awhile. And everyone else had to sit
and wait. Your poor message is last in
line so you have to wait. So finally this
message gets across the closest link. And
then the, the next message starts being
sent and you have to wait for that message
and wait and wait and wait and wait and
wait. Wait and wait and wait! Hey wait.
Okay. Now its finally your turn. So your
message finally gets to use the one
connected line. So they, they are sought
of stand in line until your turn
[inaudible] runs acrossed. And they all,
these messages aren't destined for just
one computer away, then they got to go
through the whole thing again, move across
the next link until you know eventually
you move across one link over here and
then go another, and then finally talks to
the people who get their email. So its a
sought of dedicat ed line and you had to
stand in line to get your chance. And the
key thing here is each of these lease
lines has a fixed cost 24 hours a day
seven days a week, and it's very dependent
on the distance, so we saw a weird
phenomena. ≫> And that is. ≫>
If we could add hops, it would slow our
message down, but it would reduce our cost
greatly. And so let's just say we have
Michigan State University, which is where
I got all my degrees from. University of
Michigan here in Ann Arbor, which is where
I work. And let's say, you know, we're
connecting to the rest of the world, and
we're going through Cleveland, where Case
Western Reserve is. Case Western Reserve
was the early innovator in, in networking,
and so we have two leased lines with a
certain distance, right? One from East
Lansing to Ann Arbor and one from Ann
Arbor to Cleveland and so we're sharing
the cost of these lines between three
schools and we can all kind of connect to
the rest of the internet, all connect to
the rest of the internet out here and, and
we just, some of us have, are farther
away, and so we take longer. The folks in
Cleveland are closer. Like all the rest of
the connection to like the East Coast and
the West Coast come through, say like here
Cleveland, but if we can simply convince
somebody in between us like say Toledo to
add a connection. Now of course. Of
course, this, [inaudible] Give me green.
There we go.'Course this line probably
goes, probably went around when we just
went straight to Cleveland, here. But
basically if we can convince Toledo to
sort of put in their computer and hold
onto our messages for a while, we could
send now one hop, two hops, three hops.
But the cost now is not that different,
because the original long line between Ann
Arbor and Cleveland was distance
sensitive. And, so, you can think of this
as, you can get this almost for free. And
now we have a whole additional university.
Both to send stuff to, and they can send
to the whole world as well. And so this
motivation to effectively take the same c
ost, and now basically we're taking this
cost, and dividing it by four schools. And
if you start thinking about it, it doesn't
take long to say, "You know what, let's
put one here, one here, one here, one
here, and one here." Because the cost of
the phone company isn't that different.
You can think of each of these as adding
some delay to your message. You know, and
given the fact that each of these
represents an outbound queue of messages
that are waiting to be sent, there's some
delay. There's some cost adding this, but
It's so much cheaper. So our faculty have
to wait another twenty minutes to get
their mail through if we can bring that
many more universities on. And so this
just works out. There's this sorta
motivation that if you can find an
intermediate person, geographically
intermediate school or university or
company, and you can add them in, you can
replace one long link with two short
links. And this led to long chains of
mail. And so from the mid 70s to the late
80s most academics were communicating
through a network that was like this. It
typically was email and I recall when I
first started to use national email. It
took a long time for mail to go back and
forth but it was actually quite magical I
mean who cares if it took an hour. Now we
expect it in three seconds. We send an
email and hit the refresh buttons, hurry
up, hurry up. You know, it could be hours,
it could be days if you were going far
enough and your message was long enough
and you end up behind too many queues. And
so you had this one computer locally and
every once in a while you'd do most of
your communication computation locally.
And every once in a while you would fire a
note off and that would kind of fight its
way through all those successive
connections. This is sort of the life in
the early 1980s. One of the, most widely
distributed networks of this kind was a
thing called Bitnet. And Princeton was
kinda the hub of this and these tendrils
of connections ran out from Princeton. And
by connecting to a, a network with lots of
oth er folks, then you had more people to
talk to. And the more people that you,
that were connected the cheaper that it
was for everybody. So it was a pretty, it
was the perfect kinda thing that caused
people and com-, universities to want to
work together, because together their
shared cost was much, much lower than to,
to provide this uniform connectivity and
email. So at the same time, during that
same period, where most of us were using
store and forward network, with our one on
campus computer, a bunch of computer
scientists were funded by DARPA. The
Defense Advanced Research Projects
Administration, to imagine a different
kind of network. And the idea was direct
connections are expensive. The long trails
of store and forward networks, they're
very slow, and if you had a giant message
that you got behind, then what, how do you
get past that. It could clog the system up
for, for hours, if not days. And, and how
do you keep from failures breaking the
entire system? If you think about a store
and forward network, one computer going
down would cause data to back up on both
sides of that computer until it's done.
And so, you don't really wanna have one
outage and, and how if we have sort of
instead, instead of just a few messages,
what if we just wanted all the messages to
go simultaneously, so that there's more of
a fair allocation of the network, rather
than whoever gets there first gets it all
until they're done with it. And, and so
Darpa wanted to solve the problem of
outages. You know, many will say that it
had to do with, battlefield conditions,
which is probably true They expected that
various connections would go out in, in,
in dynamic situations. Maybe it was that
stuff was moving. But also how to be more
efficient. And so, in effect, you can kind
of think of this as all a game, where the
phone companies own the wire. So everyone,
even government, even military has to
lease the wire from the phone companies.
And so everyone is like doing research to
figure out or creating systems to figure
how not to pay t he phone company so much
money, okay. So these research networks,
and so if we look for example at this one
down here by 1972 they had this network. I
have my, I, yes I got a caller. So they
have this network by 1972 and it's got,
like some [inaudible] right around twelve,
fourteen, fifteen hosts in it, and it goes
cross-country. Now, now, the, the key
about this is in 1972 to have leased lines
that were up 24 hours a day, seven days a
week, all the way across the country? Very
expensive. But hey, it's a government
project, and the government says this is
important so we're gonna spend the money
because, so we're imagining battlefield
communications of the future and our own
ability to do computations so they could
have comp-, computational equipment all
over the place. So this was very
expensive, but research dollars were being
flooded into it, because the q, they were
solving a research question. If you just
think about this as a network, it was not
all [laugh], it wasn't sorta like, it was
so costly that the average person wouldn't
like, pay $fifteen a month to use it. It
would just be that costly. But it's okay.
Now if you look at this, you see that
across the United States, there was always
at least one connection. They had three
cross country links with totally
independent cross country links, with the
ideas that you could take one of these
things out, and you could still be
functioning. So they, they were able to
research all these things right, as well
as the efficiency problem, which they
solved using packet switching. So, by the
mid 70's there was quite a few folks on
this. And for a group of people they just
started using it in production. It was
pretty cool, right? If you were, if you
were one of these universities or
companies, you had a pretty cool,
futuristic world. You could, you could
send email and get an answer back in two
minutes, or a minute, or 30 seconds, even.
And so it was kind of this futuristic
world that was heavily subsidized by the
government in the name of researching. And
so there are two essential things that
really came out of this research. And one
is the notion of what was called Packet
Switching. Packet Switching basically
eliminates the problem where once the
message starts using that leased line
wire, you have to wait till they're done.
As, as I showed in that in that store and
forward. What you want is to be able to
send little pieces. Break the messages up
into little pieces, and then they, they
Each, each message has a little bit of the
network connectivity and then the next one
comes after it. And so you could have many
messages going at the same time. And a
real long message won't fill up the
network, fill up the connection forever
and ever and ever. So it and it also
allows, if you to break the message up
into small parts, they can flow over
different paths. The other thing that they
figured out was this notion of instead of
oop, oop, oop come back here, come back
here. Instead of using computers as the
intermediate stop points, because in store
and forward you could have a lot of
messages so you tend to store them on
disks. Whereas routers, these packets were
smaller individually than the entire
message and so they didn't need to store
them nearly as long and they didn't need
as much storage. So these are, routers are
just a form of computer, right? But they
were specialized for moving just data from
one connection to the other without long
term, without storing that data for a long
time. So, I like to think of packets as
postcards, letters and think of the Packet
Switching Network as the postal system. So
let's say, for example, I had a friend,
and his name is Glen, and I want to send
him a message. I want to send him a
message that's hello there, have a nice
day. But I have a limitation. I have
limitation. All I have is postcards that
it can, that can store ten characters on
them, and I have to send my message to
Glenn using only 10-character postcards.
And so, before Glenn and I part ways, we
agree on the following protocol: that I
will take the first ten characters of the
me ssage and put them on one postcard, and
then I will put an address from Chuck to
Glenn, and I'll put a sequence number. So
that says that hey, hey Glenn, here comes
a message, this is part one. Then we take
the next ten characters. And I mark that
as part two, from Chuck to Glen. And then
here's the third part, it's marked as part
three, from Chuck to Glenn. And, so, what
can I do now? Well, I walk out to my post
office box, and I send'em, I just stack'em
in. I might stack them neatly in order.
Now, they go through the postal system.
Like, they get dropped, some get dropped
on the ground. A couple of them get lost.
Or they end up on the wrong truck. They go
through Kansas City by mistake. Blah,
blah, blah, blah, blah, blah, blah. Blah,
blah, blah, blah, blah, blah, blah, blah,
blah, blah, blah. But, you know, some days
later. They start arriving at Glen's
house. And so, Glenn goes out to his post
office box, and he gets a message. It's
hello ther-, and it's sequence number one.
So it looks like Chuck is going to send me
a message, and I've got the first part of
it. That's pretty cool. So then he goes
out the next day, and out comes, nice day.
But this is #three. So, because I've
numbered them, Glenn knows that there's
some missing bits, right? So Glenn just
can hold on to them, and leave a little
space on his kitchen table for what he
hopes to be message number two. And so
message two finally comes out. And now
Glen is capable of saying, "Looks like I
got the whole message and I can reassemble
them. And, surprise, surprise. I have just
sent him. With a lot of effort in three
packets. Hello there! Have a nice day."
And so this notion of breaking the message
into packets, labeling each packet with a
sequence number, and then sending them to
this network that can take multiple paths,
You can even have a situation where the
you know, the message would go across one
link, it would get lost and then it would,
you know go across a different link. So
you have ways of recovering. You can
recover the messages. We'll talk about
that later as well. So this ends up with a
sort of a structure that has these
computers that are specialized routers in
the middle. And the routers have multiple
connections. And if we take a campus, for
example, and the campus has some computers
and we have high-speed networking on this
campus. We have some, you know, stuff in
our offices on the campus, and then we
have some stuff in the machine room and we
talk to these things. And then, somehow,
our entire campus has a little spicket to
the outside world and this is our, sort
of, campus router and we get this router,
and then there are, sort of, intermediate
routers that are inside the network. And
if you sort of look at a router, a router
sort of simply forwards traffic and the
traffic now is these small packets, rather
than whole messages, so you don't need a
disk drive on these, on these routers.
There's no disk drive on these routers, so
that they just kind of grab a packet and
they forward it. And the systems are
trained. And the software does not
overflow the network. We'll talk about
that later, much later. And so these
routers have these real simple view of the
world, they've got some incoming traffic,
they've got some outgoing traffic,
outgoing traffic. And so they just grab
and forward. It's like a intermediate
postal spot, right? They, they grab big
thing of. Postcards and books. Send them
to the right place and, and they get where
they do and so eventually the data gets.
Getting a little sloppy, getting a little
messy, here. Eventually, the data sort of
is broken up, finds its way to the other
end, and then dumps out in some campus
local area network and then somebody sees
the data on the far end, okay? And so it
might different, take different routes,
you know? It might get lost that might
crash and then it has to get sent again on
a different route. And so these things,
these little pieces, these little
postcards, find their way through the
series of routers. And we can, we both can
see sort of like a, a whole campus being
connected. We can see individual folks who
are, buying, buying some dial up through
cable or DSL, and at some point we like to
represent this whole thing. Here is this
big cloud, this you don't worry about the
detail inside here. Call that the cloud.
We'll see it in the future slides it`s
just a cloud, a white, fluffy cloud. That
means that we are trying to hide the
detail. But in there it`s just a bunch of
things that are connected. In a way it`s
not that different in the store and
forward network, except for the fact that
every message is tiny, so it doesn't clog
the whole network up, which means that
routers don`t have to have a lot of
intermediate storage to hold on to these
packets in flight. And it also means that
every packet can take a different path and
if things get loaded up, they can
dynamically move. And so. Here's just sort
of an example problem to solve. If you
think about it, these routers have a very
limited view of the world. And there are
hundreds of thousands of routers around
this world right now. And they don't know
the entire network, they kind of know the
lines that come in to them and the lines
that go out, just like a post office in
Kansas city doesn't know every address,
every house in the world. It just knows
the trucks that are coming in and the
trucks going out. And so these packets
that have to and from addresses can get a
little confused at times. So we won't
solve this but if, if we had a situation
where This particular packet would come
into a router, and it would route it here,
and then this packet would see it and then
it would route it this way, this packet
would see it and route it this way, this
packet would see that this router would
see it again and say, oh, I gotta route it
that way. And so we end up in this
situation where we would create a loop.
Okay. So this is the kind of technical
things they had to solve to keep these
things from going round and round and
round and sort of melting the network.
We'll talk more about that in a bit. So.
This was DARPANET. It was doing research
on these kin ds of problems. The kinds of
problems of, you know what's the best way
to do this? How big should packets be?
What should, how long should we wait until
we send a packet again? You know, this
kind of thing. And so that was our
research network. And the, that could've
gone on forever, it might've been a purely
military project, but. At the University
of Illinois - Urbana Champagne - folks
started to think about super computers and
starting all the way back to Bletchley
Park, science was enhanced by the use of
computations. And so as the 70's and 80's
were happening, all these scientists were
sort of like, "Wow, I can do better
physics. I can do better chemistry. I can
do better material science. I can invent
new plastics. I can do pharmacy. I can do
all kinds of things. With computers. And
so what happened was is everyone started
asking the government. For money. For
computers. It's like, "I need a bigger
computer. And if I, if I had this bigger
computer. I could do research." Matter of
fact. I was part of all this. Matter of
fact I wrote a book. High Performance
Computing. Here's the book. That's kind of
what I did before I became an internet
guy. These are beautiful things. Here's,
this isn't, was my baby, I never got this.
This is like about $8,000,000, it's not
small like this, this is a model of a
Convex C3800 supercomputer. And each of
these was the size of a refrigerator, it's
slightly taller than me. I would be about
this tall, right here. And each of these,
I think this is like, like I said, like
$8,000,000 or something. And I wanted one
just for me. And so the problem is, is
that, you know, I'm a nice guy, and I'm
probably worth $8,000,000 of the
governments money without a doubt, but not
that the government didn't always think
about that. So we couldn't all have out
own personal computer, or at least our own
personal supercomputers. Today, of course.
This has about as much power as this, but
this is not a history of computers.
Computation. The iPhone is as powerful as
this thing, it literally with abou t as
much storage, But what happened was, is,
all these scientists would say give me,
give me this supercomputer. I need a
supercomputer to do this, I need a
supercomputer to do that. And the National
Science Foundation said oh, hmm, well, why
don't we just buy a few of these
supercomputers and put them in these
supercomputer centers and then let people
connect to them. And then make people, and
make it so they could share, so we didn't
have to give every single scientist one of
these things. And so. The notion that we
would create a network to connect these
things, again, seems completely logical
today, but in 1981, 1982, 1983, it wasn't
entirely the most logical idea. And of
course, the telephone companies might have
something to say about that and so the
next person that you're going to meet is
Larry Smarr from NCSA, the National Center
for Supercomputing Applications. And Larry
Smarr was one of the early innovators that
sort of realized that we had to build
computational infrastructure and internet
computational infrastructure. And did a
lot of work to convince the federal
government that this is something that we
should do. And so let's go ahead and meet
Larry Smarr.