Oh, oh hi, sorry. I was in, I was in the forums.

[SOUND] I guess it's time for this weeks video.

So this week we're moving from history to technology.

And in a sense were going to tell the same story again.

you know, we'll sort of start from the beginning, but we're going to start from

the bottom and move upwards so it sounds as though we're kind of taking about the

same things, but just from a more technical perspective.

Don't worry, you'll be fine. we're not going to over, there's no math,

I promise, there's no programming. Don't worry about that.

So of course, I appreciate IEEE Computer Magazine, who I write for to let me use

the articles that are associated with some of the videos, as well as the folks

from up in Michigan who have helped me, so.

Hang on a sec. Now hang on, I've got a phone call.

I'll be right back. Okay, well that's been sent.

Sorry for that interruption. again I thank everybody for the use of

the copyright materials. part of my hidden agenda in this class is

that you will be able to look at this XKCD comic and you'll understand the

humor. So, I'm going to stop for a moment and

let you look at it and see if you get the humor.

Okay so maybe you did, maybe you didn't. But hopefully by the time we're done,

you'll have a better chance of getting here.

So if you recall, the most academics throughout the 60s 70s and 80s.

The best they had access to was this store and forward networking where you

would send a message and it would go into a computer and it might sit for some

time. Then it would find its way across a

network and it would sit for a little while longer.

Find, and, and hop, and there are these multiple hops.

We call these hops, and each of these would be a couple hundred miles perhaps.

And to go from say Michigan to Stanford, there might be fifteen or twenty hops.

Hops are as much, defined by geography where we're all trying to optimized the

cost of these long distance connections. So the thing that really characterized

store and forward networking was that there wasn't a lot of sharing.

One message was being sent across one of these links at a time and all the other

messages just kind of waited in time, in line.

Now, that the, the, the innovation that happened in the research networks in the

60's through the 80's ultimately unbodied in the ARPAnet was the fact that they

were going to be packet networks. First, how do we share one link, so that

a long message doesn't clog it up? And how do we deal with outages more

dynamically? With the idea that, eventually, you could

send a packet across the country. And back into a computer and back out in

maybe a half of a second. And that was seen as something to do

rather than. Somewhere between ten minutes and a day

or two. So how to do this efficiently.

It was a research project. And it was a research project that lasted

almost 20 year-, well, 20 years. And, 20 or more years.

And, and it, one of the neatest things about this is they were able to throw it

away and rewrite it a bunch of times. So that they could put something in

production, see its flaws, see what was good about it, and they had the money

from Darpa to throw it away and rewrite it.

Now by the late 1970s it had 100ish computers on it, here's a picture, and

these are not the schools, these are the computers, these are.

You know, listing all of the computers that are there.

And so the engine, the, the innovation that they spent 20 years perfecting, was

the notion of packet-switching. So how can we simply move data

simultaneously across the connection? And the essential brilliant idea Is break

the message into packets. And again to sort of review, here I have

a message and I have post cards that only can handle 10 characters.

And i'm going to send this "Hello there have a nice day" message to Daphne in

California. And so I have 3 10 character post cards.

I basically put a from address and a to address on each postcard and a sequence

number on each postcard, 1, 2 and 3, and I put ten characters on each and I stick

them in my mailbox. I just put them in.

Put the sign up and then I wait. Now let's just say you know, that the

post office person comes and picks this first one up and it goes to Chicago from

Michigan than it goes to Omaha then it goes to Denver then the second one kind

of falls on the floor and it ends up going to Charlotte North Carolina.

And then they're like, Why is this here? And then they send it to Atlanta to get

checked. And then they make sure.

And, and this one goes and. And Number 3 ends up going to St.

Louis. And then Tulsa.

And then Colorado. And this one gets routed back to

Charlotte because they weren't sure why they sent it to them in the first place.

And then they send it to Memphis, Tennessee and then this one makes it to

California. And this other one sort of goes to

Phoenix and this one finally gets to Dallas, Texas.

And this other one goes finally. And the third one goes to Tucson,

Arizona, then Las Vegas Nevada and finally he makes it.

So, these poor little postcards I put in They have each a different journey.

Right? It's a little adventure for each of the

postcards that they go through. And, so, we don't, I don't see how this

works on my end. I put in three postcards.

I numbered them, one, two, three. But on the far end, in California, Daphne

opens her mailbox. And has scribbling all over her mailbox.

Oh, the shame! No, actually we'll get rid of the

scribbling. So Daphne opens up her mailbox, and she

doesn't know how these packets got there, I mean, postcards, how they got there,

out comes the first one. She goes, oh looks like I just got a

message from Chuck, but I only got part of it.

Then out comes the second one shortly thereafter.

Then after a long period of time, finally the third one comes out, but it's

actually the second one. And now she understands that she's got

the entire message, and she reassembles it, and away we go.

And so this is the basic notion of how all those postcards can share the

infrastructure. All, they can all be invited at the same

time. They can take different paths.

You don't have to connect them all together like trains, basically.

and, and this led to a shared network infrastructure.

And so the computers that were in a store and forward network, went from, sort of,

big powerful computers with disk drives to really tiny computers with a single

purpose of forwarding packets. Rather than longs term storage of

message. And so, so the, the store in forward had,

longs term storage of messages, in the routers.

So the, but, I mean, not in the routers. In long, in the store in forward, the old

one, long term storage was somewhere in the network.

But this is only short term, right, short-term storage, short-term storage.

And so when a packet comes out, it simply has to find its way through a series of

hops. It still hops and it's still connections,

but basically it has to find its way across the Internet.

And then we would take whole campuses Like University of Michigan, say, or

Stanford for example. And, then we'd have sort of various kinds

of local networks on those campuses, and computers on the campuses, and servers

and various things. And they would route all their packets,

all their data out to the Internet, and then the packets would find their way

across the Internet, and then we might have a home connection as well that might

connect. And so, the shared, shared Network

infrastructure focuses only on packets, not reliability or anything else.

[COUGH] So, this notion of hops didn't go away.

As a matter of fact, I don't know how many hops it took to get from Ann Arbor

to Palo Alto in using BITNET. I'm going to guess it might have been 16

or 20, in these days it takes 16 or 20 hops.

The difference is in the internet, the 16 or 20 hops happens in a hundredth of a

second, or a tenth of a second. So the ho-, notion of hops, and the

notion of intermediate computers is still present in the internet, and the TCPIP

networks that we use today. And so your, your, your message sort of

leaves you the host that you're in, hops to the first router, and then hops to

however many routers. Dot dot dot, this is more of a dot dot

dot thing. And then finally hops its way out to the

far host. And so this is sort of either your

computer, maybe Stanford's web server, and then there's this series of routers

in the middle that it hops through. Okay?

And so, the problem of what data goes between here and here is you gotta solve

a lot of problems, right? How to, all, all kinds of problems, and

so, in order to simplify the solution or break the solution into simpler, more

manageable parts. They came up with a layered network

model. Now this is a cartoon and computer people

love drawing cartoons and saying this is our architecture.

This is our frame work. This is our approach.

Sometimes they're helpful. Sometimes they're not helpful.

Sometimes the cartoon is just kind of a cartoon.

But what's usually being communicated in these kinds of pictures is they're taking

a big problem and breaking it down into some subset, some set of smaller

problems. So, the whole problem they've gotta solve

of getting data reliability reliably across the whole country is that big and

if we can break into four pieces. And work separately and come up with ways

to let these pieces interact, work separately with each onem then maybe

we'll have a better solution, okay? So take a problem that's so large and so

complex that we might not be able to solve it and break it into four smaller

problems Gives us a better chance of solving the four problems.

Now, there's a certain art to picking how you break the problem up.

And there's more than one network model. The one that we use in the Internet is

the TCP/IP or the Internet Protocol Suite model, and we'll meet some of the people

who designed that. there was also a model called the 7 layer

OSI model, Open System Interconnection model.

So there's a model out there that has 1, 2, 3, 4, 5, 6, 7 layers.

And they all have names, and they all have purposes, and they all have

definitions. this was not quite as po-, this was not

very popular. And I doubt that it's many places.

because the TCPI model is the one that kind of won.

Um, [COUGH] and so that's the one that we're going to study.

But that doesn't mean that, that has to be the only one.

But it is the one that has become popular on the internet.

So, once you break the problem down from a big problem into four small problems,

you have to develop documents about how these layers work together.

How various computers work together. How routers work.

All these things and so early in the process of the development of the Arpanet

they created an open process to build these specifications.

Where they would vite, invite engineers from all kinds of companies and

universities and who knows what any expertise to show up at a meeting.

Several times a year. I mention the IETF meeting when I was

talking about the history of the Internet and Tim Berners-Lee boff, that only 15

people showed up to. That's this meeting, the Internet

Engineering Task Force, which is just a bunch of engineers that get into rooms,

many rooms, and solve a bunch of problems.

And so the standards that come out of this, you can go read them, they're all

open documents. They're very throwback documents.

They've got this text thing that's dial this like, beautiful upper case, it

reminds me of back when printers didn't have a lower case.

and these are documents that describe how most components of the internet work

together. So, the layered architecture that I just

described where you have a computer and then it sends data out to back of it.

And then hop, hop, hop, hop, hop through routers, and then the data arrives at the

destination computer, And then it, the destination computer sends it back it

hop, hop, hop, hop, hop coming back and it comes to you.

We're going to expand on this a little bit.

So each of these hosts on the two ends is going to expand to this much right here.

So that's one host. That's another host.

And each of these routers is going to expand into this.

So this bottom picture is just an expanded version of this upper picture.

So that we can see how the Internet layers work together.

So when you send a message in on your computer, it's -- this is all software,

application, transport, Internet link. So this is all software, and then it all

comes out the little plug on the back of your computer.

Right. [SOUND] So, maybe your computer has a

little plug like this on the back of it. Right?

That's an ethernet plug and here's an ethernet wire and I plug my local area

network in. Right?

And so, this right here is that right there.

That's kind of the plug. It might be WiFi which means it's kind of

like air but If you have a wire like this, then it's a little more tangible,

and there's something on the other end of that wire.

This is a very short wire, but there would be a router on the other end of

that wire. [COUGH] The router takes the data off the

wire and then forwards it just like a post office, intermediate post office

would do, onto the next link. On to the next one.

And this would be multiple hops in here, they're only showing 2 hops.

And finally it's on the last link going to that computer at Stanford.

And then it goes up through these 4 layers of software and then goes and does

whatever it's going to do. Now when they send you the response back

it kind of goes the other way. And back up and back to you.

So if this was you in a web browser, you know and then here's me.

And here's a slide I've got, right. This web browser and you moving your

cursor in the web browser is sending stuff back and forth to the [UNKNOWN]

servers right here. And that's what's happening, alright.

So If as you look at this picture over here, this left colmn is the host you're

coming from. The right column is the host you're going

to, and the riders are these things in the middle, okay?

Okay. So let's start looking at all four

layers. We are going to start at the bottom, at

the link layer, then we're going to move up.

So if you recall when I talked about the, the, the basic reason why we even have a

layered architecture at all, is to simplify the problem.

To simplify the problem. So the idea of the link layer.

Is, the link layer is a, it, it on-, only worries about getting the data across one

hump, right? Only worries about getting the data from,

across one piece of wire. White like voltage goes on these little

wires, and. How we send the stuff and if more than

one computer's using the same wire, how do we share?

It's a complex enough problem but, we don't worry about the whole world.

We worry about this one link layer and so the link layer is sort of like the

connection out of one computer and into another computer.

And, and this might be fiber optic, might be, you know, 40 miles or it might be 40

feet for all we know. But it's one link, okay?

And then, a router pulls it off that link and then forwards it onto another link.

If you go to the post office, you could think of the, the person who picks your

mail up from your house, with the thing on their shoulder, that's one link And

they put it in a truck which takes it to a place that puts it in a semi truck.

And then it puts it to another place that puts it on a train.

And the semi truck and the train and the postman or postwoman are these little

link layers. Each person or semi truck is not taking

it all the way, they're just getting a little farther.

So that's the link layer So the link layer doesn't worry about the rest of

this stuff. It really, the stuff that is defined on

this wire, it doesn't even care if there's a world wide web or anything.

Its job is to get data across one foot. That's what its job is.

Is it up, is it down, how do we share? It doesn't care.

It's got a very narrow view. So.

What it means is we can zoom in on this problem, right.

We can zoom in and focus and ignore everything else.

So the link layer basically asks questions like, you know, I've got some

data inside the computer and I want to send it out.

How do I encapsulate it? What if this is shared?

How do I deal with that? And common Link technologies that we see

are like Ethernet or WiFi or cable modem, DSL, satellite or optical.

These are all links layers of one form or another.

So if you're going back to this Ethernet link layer, which is one of our favorites

cause it's kind of ubiquitous. Other than wireless it is probably

Ethernet is the most ubiquitous link layer.

[SOUND] So, when the manufacturer builds an ethernet or a wireless adapter, they

actually build into it, which is probably right here.

No, it could be right here, they build into it a serial number.

So, they, they mark this as having a serial number.

And you can actually whether you're on Windows or Mac, you can work your way

down to find the serial number for your actual manufacturer piece of equipment.

So there's mine on the Mac. That's an example one on the Mac.

And that's an example one on the PC. They tend to be six two digit numbers

concatonated together often with colons or in this case with Windows it shows

dashes. And so this is a raspberry pie and all

the raspberry pies are going to actually have similar prefixes and then serial

numbers within that prefix. And so the raspberry pies get

manufactured, they come out of the manufacturing line with this sequence

number just kind of going chug chug chug, up up up up up.

And, then they get shipped to all corners of the world.

And so, these numbers are not the numbers for this to get across the world, all the

number, all these serial numbers at the link layer are good for are to get across

one connection, in case that connection is shared.

And if you were to plug this into a hub, and you're plugging other computers into

that same hub. Well then you're sharing that connection.

And this computer might see the traffic for that other computer.

And so we use the physical addresses so that this computer knows which of the

packets belong to it on the piece of wire.

So, wired ethernet or wifi certainly if you're sitting in a room with a bunch of

wifi computers. You're sharing the air.

So let's come up with rules how to share this, how to share nicely and how to

behave. So the way it works, ccc, is you

could...the ethernet could easily have a bunch of computers hooked to it right.

Cause its really shared, and so they're all sitting here, they're all talking

simultaneously. No.

But, we have a com, we have a pair of computers that want to talk.

So, it's not just one. It's got a bunch of folks sitting here.

[SOUND] Just draw a bunch of em. Bunches of folks.

They're all talking or all potentially talking [SOUND] and we gotta figure out

how in the shared medium. Like this is, think of this as a hub and

they're all connected into this hub including the two we want to talk.

They're all connected in the hub, including the two that we want to talk.

And how can these two make sure that the data goes, is, is the data that's sent

with the intention of going to this computer, this router.

How can it make sure that it gets there? Well, if it knows the address of the

sending and receiving unit It just encodes that, in the packet.

And that way, it sort of goes by all these folks.

Let me change colors here. The packet we send goes by all these

folks, and even if they all heard every one of the packets, they all know that it

doesn't belong to them because they all have a different number and this is the

only one that responds to it. So that allows them to share the wireless

or the wire, and, share the wireless to the wire and have it all work out for

them. So, [COUGH] so that's the idea, the link

layer. Now, again, remember this might only be

fifty feet, or fifty yards, or fifty meters.

And they can be meters. It's only 50 meters, but you still have

to share the network in that 50 meters. And then your router takes it off of that

and forwards it on a different link. And that's how we get across the country.

But that's the next layer up, and let's not worry about that for now.

So, the idea is think about one link, whether it's fiber optic or cable, or

wireless, or wifi. And so what are the kind of problems that

you have to solve on this link layer? Well, so one of the things that's cool

about many link layers is, like wireless and ethernet, is they can be shared, and

it makes it real easy just plug new computers in.

So you just sort of like, here's a hub, and you just plug another computer in,

and it's on the network. But they have to come up with a way to

avoid the chaos when they're sharing. And so, the way Ethernet does this is

with a technique called, Carrier Sense Media Access with Collision Detection.

And as you'll see, a lot of things we do on the internet have a lot to do with

courtesy. Meaning that we're just nice.

And if everybody's nice, it all works out.

kind of like driving in a car. Just everyone can't run through the stop

sign at the same time. Some of you gotta stop, got a green

light, your turn, my turn, whatever. This probably wouldn't work so well with

cars, because sometimes you do have a collision of packets.

You just have to wait. Packet collisions don't crush your car.

They just slow your data down. So, here we go.

This is basically carrier sense media access with collision detection.

The first thing you do if you want to send some data, say we got some data

inside our little computer and we want to send it out, knowing that there might be

other data going by The first thing we do is listen.

We listen to what's on there. If it's not silent we just wait until it

goes silent and then we start sending. So if someone's already using it, why

crash into their data and crush their data?

Because there's, it's shared, there's only one.

You can't both be sending at the same time.

So first you listen. Once it's silent you begin transmitting

data. [COUGH] And then what you do is you also

listen to your own data and if your own data is coming back, then it's sounding

pretty good. and And then if, if you.

If there is a collision, because there's a chance that two computers will want to

send at the same time and they'll collide.

Then what they have to do is back off.. And they have a sophisticated random

number calculation, so that they back off not always the same amount.

Each computer backs off a different amount, and they make it so that it's

fair. So that one computer's not always backing

the most off. So it's very fair.

So it's, when you detect a collision, you retransmit after a random backup.

First, you avoid collisions. But in the, in the rare case that you

have a collision, then you do this. And so,

[NOISE] I want to introduce you to the person who invented Ethernet, Ethernet.

Hmm, This is Robert Melcalf and he was working at Xerox, Palo Alto research

center, PARC, everyone calls it PARC. And they had, they were building what is

most considered the first computer, the Alto computers.

They were connecting line printers, or, fast printers to these.

Pate, Xerox was building the first, laser printers.

And they needed something faster. And so they just started building, and

they built this thing called Ethernet. Now it's a little different than the

Ethernet that we see, the first Ethernet that they built is a little different,

you can see a picture of it down here. If you ever go to Xerox Park and get in,

you'll go see this little museum they have.

it used a single piece of cable that ran along, they just ran it down the hallway,

and they would connect a tap in And then they would use that to send the data.

So it would go down the hallway and a tap come out at each office, and they were

truly sharing the media. This comment is not shared, except when

you plug it into a hub. That's what makes the shared.

it was inspired by an earlier wireless network where they were doing not

collision detection, but retransmission called Aloha, and a lot of early network,

packet network inspiration comes from the Alohanet in Uni-, comes from University

of Hawaii. And because they didn't have anything

except wireless, but they did have wireless, they did a lot of really cool

early research on wireless. And so let's go ahead and meet Bob

Metcalfe the inventor, one of the inventors of ethernet.

Introduction / The Link Layer

互联网的历史、技术和安全

与讲师见面