UMass Professor Explains the Internet in 5 Levels of Difficulty

Hi, I'm Jim Kurose,

I'm a professor

at the University of Massachusetts at Amherst,

and I've been challenged to describe the internet

in five levels of increasing difficulty.

The internet is the most technically complex system

that humanity has ever built.

The internet is a network of networks.

It's a platform on which all of the internet applications

that you've heard of can be built.

[bright music]

Hi, it's really, really nice to meet you.

What's your name?

My name is Skylar.

Skylar, we're here to talk about the internet,

and I bet you must use the internet a lot, right?

Yeah.

What's your conception about what the internet is?

The internet?

For me, it's just something to use when I need

like to search up something or watch videos.

The internet is, physically, these computers

that all talk to each other.

Billions of computers, in the case of the internet.

The internet allows us to do

a lot of really, really interesting,

what we call applications.

You ever think about how that video gets to you

over the internet?

Yeah, I have no idea.

Got a favorite movie?

Matilda.

Matilda. All right.

We're gonna actually build an internet.

I've got a couple of things here that I wanna show you,

or a couple of toys, actually.

Okay, let's pretend that these round balls are computers.

And the internet is something that connects them.

And right now, the internet is just one communication link.

And Matilda is sent over the internet from this computer

to your computer.

So the internet is a network for carrying information

from one computer to another.

Now this network here looks pretty simple, doesn't it?

Right? It's just one thing.

Should we add some more friends in?

Yeah.

Let's say we want to get a video from here, over to here.

How do you think that video would sort of travel

through this network?

Maybe it could go to here, to here, to here, to here.

That's right.

So that's pretty cool.

There are actually lots of different ways to actually go

through the internet to get from what we call a source,

the place that's sending the information,

to the receiver, the place that's actually gathering

the information together.

And that's something we actually call routing.

Huh, but wouldn't it just be easier

for it to go from here to here,

instead of going from here to here,

to here to here?

Yeah. So that's a really good observation.

In most pieces of the internet,

that's exactly what would happen.

We want to take what's called a shortest path.

But still, there are multiple paths.

And why do you think that might be valuable?

Maybe one way is messed up or broken.

So you go the other way.

Exactly.

So, Skylar, that was a great discussion

about what we just built.

And I wanted to talk to you about,

or ask you about maybe one other really important

part about networks.

And it's not so much the thing itself,

the physical thing,

but more about the rules about communication.

That's governed by something that are called protocols.

Are you up for one?

Yeah. Knock, knock.

Who's there?

Lettuce. Lettuce who?

Lettuce go on.

[Skylar and Jim laughing]

A knock, knock joke is an example of a protocol, right?

The computer that you are using say, makes a request,

you ask for something, you get something in return.

In the internet, there are protocols everywhere.

So that two computers that have never talked

to each other before know the rules

for talking to each other.

This global internet with billions of people using it

are just lots of smaller networks

that are all hooked together to each other.

But also, what the internet allows

are all of these what we call applications, Zoom,

video playing services,

can all run on top of the same internet.

Yeah, so there's one internet for all of 'em.

Exactly.

There's one internet and lots and lots and lots of things

that you can do on top of it.

[bright music]

So you're a student in high school, is that right?

Yes, I'm a sophomore.

Well, we're gonna be talking about computers here today,

and we're gonna be talking about the internet.

I always like to think of the internet by analogy

to say road systems for example,

where you have roads in your neighborhood.

You have state roads,

you have the Interstate Highway System.

And so the internet is a lot like that.

It's an interconnection of local roads,

local networks like the network in your house for example.

How does like all of the networks in my house connect

to all the city networks?

Wow. Great question.

Often, it's a little blue wire called an ethernet cable.

So that cable is able to bring bits of information

up into your apartment at say, a billion bits per second.

That's pretty fast, right?

Literally a wire that goes between a box in your apartment,

sometimes called a router or a modem in your apartment

that comes from an internet service provider

come into this first network and then that network connects

to another network connects to another network

connects to another network.

You could FaceTime somebody who's like in Australia.

You can talk to them at the same time,

and like you're reaching the same signals.

So how is it that it gets there so fast?

We could talk about that by analogy to a road system.

It's not just one big, super highway.

It's a lot of smaller super highways

that are all interconnected.

And those interchanges are what are called routers.

That's where the links come together.

You're talking about talking to a friend in Australia.

So oh, it's coming in from the East Coast

of the United States to this router,

and it's going out say, that routers in San Francisco,

it's going out on an underseas cable over to Australia

rather than in this direction up to Japan.

So there is an underseas cable?

The underseas cables are so cool!

They're these big cables that are laid down by switches.

They cross both the Atlantic, the Pacific, the Indian Ocean.

So the undersea cables are how the networks in Europe,

United States, Asia are all connected together.

How do you connect wirelessly?

That's really what we call the first hop.

It's like from your phone, from your tablet,

from the computer that you're on,

there's no cables coming in.

You go over a wireless connection.

Wi-Fi is the protocol that allows your computer to talk

to that first hop router over a wireless communication link.

And I was wondering how there's so many different movies

or TV shows that you can download and they're all there.

And if you just play it, it just knows what to play.

Like they're all in one spot.

Ah, you said they're all in one spot.

In fact, they're in lots of spots in Netflix.

And so most applications would like to connect you

with a server that's close to you.

Server is really just a big computer with a lot of memory,

a lot of discs that store all the Netflix movies,

and also so that you don't have to cross over

too many internet links to get from where the server is

to the TV or the device in your home.

So when I'm watching Vampire Diaries in my house,

how does it know exactly what to do

without getting scrambled up?

Ah, another great question.

There's a couple of things that could happen

inside the internet.

Information is sent in these little packets of information

from the Netflix server to your display device.

And literally, each packet that arrives says,

This is the first packet for Jenna.

This is the second.

This is the third.

This is the fifth.

This is the fourth.

And they're reordered for you.

Matter of fact, your computer will say,

using the TCP protocol to the server,

Hey, I didn't get packet four, can you resend it again?

And again, the Netflix server is very happy

to send you packet four again.

The other is the internet protocol.

If you think about sending letters

through the US Postal Service,

how you've got an address on it.

So every packet that flows from the Netflix server to you

has an address on it.

It says, This is going to Jenna.

It's going to the what's called

the Internet Protocol address of your device.

Think of all the range of devices

that are hooked up to the internet.

It's totally amazing, right?

Every single one of them has one thing in common,

and that is they speak the IP protocol,

the Internet Protocol.

That was a great question.

[upbeat music]

So tell me a little bit about yourself?

I am a senior at New York University.

I study computer science.

Have you taken any courses on the internet

or studied it at all?

I've taken Applied Internet Technology.

So we've talked about backend/frontend frameworks

and libraries, things like that.

Okay, so really at the application level?

At the application level, for sure.

I wanted to ask you a little bit about what you know

about the history of the internet.

Have you heard of ARPANET, for example?

I have not heard of ARPANET.

Okay, back into the 1960s, there was a research agency

in the United States called DARPA,

the Defense Advanced Research Projects Agency.

Actually, it was called ARPA at the time.

They wanted to build this notion

of a packet-switching network.

Not a circuit switch network like a phone network

where you get a dedicated path

and a dedicated set of bandwidth and links

from source to destination.

So what would packet switching enable?

I'm sure there's something big here, for sure.

There's a lot big, right?

And so remember, this was a Department of Defense,

was they wanted to have forms of of communication

that were very robust, that were survivable.

Packets could all find their own ways,

be routed differently through the network.

So if parts of the network failed,

you could route around failures.

Sounds like the reason

for like a request response type of structure.

So you can sort of see how the network architecture

that wasn't designed to be 100% reliable

inside the core of the network,

and had that complexity built into the edges of the network.

And to me, the really cool thing is you

put this infrastructure in place,

and then all these super creative people

think about amazing things to build on top of it.

And you see this proliferation of amazing applications.

Abstraction, I think it's the reason why everything.

Ah ha! Spoken like a real computer scientist, right?

You're a computer scientist. I'm a computer scientist.

We talk about APIs, application programming interfaces.

The API for the internet is something

called a socket.

And a socket simply says,

I can communicate if I know your internet address,

you know, 128.119.40.186,

that number is the IP address of my server,

the University of Massachusetts.

If you know that, you can write a program

anywhere in the world and send a message,

and it'll pop out at my end.

I will be remembering that.

[Jim laughs]

I've heard that there are like seven keys to the internet,

something like that.

Okay, well I don't know about the number seven,

but there's something in the internet

that's sort of similar to that.

It's called the Domain Name System.

The DNS's role is to translate names

like gaia.cs.umass.edu, or ibm.com, or facebook.com

to an IP address so that your application

can actually send a message to that name,

to that named service.

This whatever quantity of people

is able to have some form of control?

So that's a great question.

Who do you think controls the internet?

I'm pretty sure the internet is fairly decentralized.

Okay. What does that mean?

No one authority holds control

over any sort of decisions or destinations.

That's 98% true.

And if you own a network, like you're att.com,

or your verizon.com, you can do, within that network,

you can do what you want, right?

So in that sense, the internet is very decentralized,

that the control of the network is up

to whoever owns the network.

The 2% where you said there's nobody in control,

there's a a little bit of centralized control.

There's an organization called

the Internet Corporation for Assigned Names and Numbers.

Its responsibility is to handle, as the name ICANN suggests,

names and numbers.

It's that little bit of centralization,

central authority that you need.

When can we see the next tenfold increase

in in Wi-Fi speed?

In terms of tenfold speeds of increases,

depending on what device you're using right now,

it's available, all you need to do is upgrade.

So the Wi-Fi protocol's called 802.11.

And this is sometimes a source of confusion for people.

How can it be that I've got a connection

at 100 megabits per second from our TV into our router?

100 megabits per second not enough?

Packets dropping?

Where do they get dropped, do you think?

Somewhere in their travel process.

Exactly, right.

And maybe they're dropped in your apartment,

but much more likely, they're dropped because of congestion

somewhere between the Hulu or the Netflix

or the Disney server, if you're watching a video,

and your home.

So even though you've got 200 megabits per second

on that last hop, you don't have 200 megabits per second

from the server into your apartment.

I see.

I'm curious, has our conversation

sort of changed your view or sort of taught you new things

about the internet?

I think that I've sort of realized

that the internet is a technology that's dependent

upon so many other factors.

Some more in our control, some less.

[bright music]

Tell us a little bit about yourself?

I'm Caspar Lant.

I'm a PhD student at Columbia University

under Henning Schulzrinne's tutelage.

Oh, good pronunciation. [laughing]

Thank you.

I'm interested in networking, IoT,

and sort of what kind of data science you can use

with the datasets that you get from such devices.

One of the things that I designed before,

starting my PhD with Henning,

was a IoT pill dispenser, essentially,

which pairs with your smartphone,

which does facial detection

and other computer vision controls

and can basically tell who's taking

some sensitive medication

and make sure that they've taken it correctly.

We have these low-power devices

they're sort of at the edge.

Is it just connecting them in across a wireless link?

Is that the primary challenge or?

Well, I think the primary challenge is that for sure,

but then an additional challenge

is keeping everything configured in the way

that you expect it to be configured.

So for example, most IoT devices require you,

when you're configuring them for you

to enter some kind of captive login portal

where you connect to a local network

that the IoT device produces,

and then you can input your Wi-Fi SSID and password.

But then say if you were to change the password

or the name of your Wi-Fi network

or you move to a new place, then suddenly,

everything needs to be reconfigured.

'Cause that's a problem that scales linearly.

That you don't want the complexity of managing them

to go up linearly with that.

You'd like it to still stay pretty flat as you start adding.

Right, exactly.

I mean, the good thing about IoT devices

is that they tend to transmit

very, very small amounts of data.

We're used to ethernet cables

that can handle many hundreds of gigabits per second

over a wired device.

What are the typical data rates for IoT devices?

I mean, not hundreds of gigabits.

No, I mean I would imagine upper bound, KB per second,

lower bound, you could see bytes per second just on average.

But I mean, say that you have a temperature sensor running

off of your Arduino that's reporting the temperature

in your house every minute.

That's going to be far less

than kilobytes per second on average.

My sense is you're spot on,

that they'll produce over time, a lot of data.

And that a lot of IoT is about computing on that data.

That computation happened mostly at the edge,

or somehow a combination between the edge

and something happening in a far away data center.

Well, my sense is right now that all that data tends

to be centralized because IoT devices

are usually the commercial products of companies.

Do you think they'll share it?

Not without some persuasion,

but I agree that these data

have massive, massive research value.

Something I'm interested in with my research

is collaborating with people who manage

these distributed sensor devices,

and then taking advantage of those datasets

and comparing them to, say you were interested

in doing a research project on how daily rush hour traffic

impacts the acoustic landscape of New York City.

Figuring out, look, this street next to this school

is causing visible ratings above what we mandate.

And so there needs to be an intervention here.

I think for a long time,

the internet hasn't grappled with,

but now has with IoT and also with cellular networks,

generally is the question of mobility.

Do you imagine in the future that it might be possible

for mobile devices not to always have to connect

through the same provider to go from one network to another?

Definitely.

I mean, we're already seeing long range networks

like LoRa that can, first of all, provide access

over a much larger coverage area,

but then also look the same

because they're set up to the same specification,

regardless of where the individual gateway is.

[bright music]

So hey, Jen, it's great to see you again.

Good to see you, Jim.

We're in level five now.

So you're the expert-expert.

I'm a huge fan of the work that you did in RCP,

the Routing Control Platform being a precursor

to software-defined networking

and the notion that rather than having protocols

actually always compute things,

that we could compute things in data centers.

I'd be interested if you could sort of just reflect

back on that time and sort of the beginnings

of SDN and where it's come since then.

Yeah, and when we were at AT&T,

the thing we found most frustrating

is AT&T would buy routers,

and they would come pre-baked with a set of protocols

instead of knobs that you could turn if you wanted

to influence how the traffic flowed,

and a set of dials you could read to understand

what was going on inside the network.

Right, you couldn't directly do what you wanted to.

Exactly.

And so we started thinking about earlier work that was done

in the telephony network, the old telephone network.

And there, they had the same problem.

And they had the idea of having a computer running a program

tell a distributed set of telephony switches what to do.

But the idea was like, wow, it was kind of a revelation,

like what would that look like if we did that?

Not for the whole internet, but at least AT&T's part

of the internet.

So in other words, use software

instead of distributed protocols

to to tell the network what to do.

Yeah, do you see the softwarization of the internet

as a whole happening?

So, so far, it hasn't very much.

I mean, basically, software-defined networking exists,

let's say within a single provider backbone,

or a single cloud provider's network or a single campus.

There's been some work on doing it at the juncture points

between a pair of networks.

But one other trend that's happening

that makes it more possible is it used

to be that to get from one end to the internet to the other,

you have access networks getting much closer to, say Google

or Microsoft or other large cloud providers,

where even, you might only go through two networks

Right, so some people have called

that the flattening of the internet, right?

I think it used to be on average,

you would go through 10 different networks

to get from a source to a destination.

Right, exactly.

And if you take that even further,

they're starting to be more edge computing

where you might imagine you might have a cell tower

connected to a small number of routers,

connected directly to a server

that's gonna be running the application.

In that case, the entire infrastructure might be controlled

by a single party.

It's totally fascinating to me

that we have such an important global infrastructure,

and yet, the laws that that govern it tend

to be very, very local.

There are tens of thousands

of separately administered networks,

and of course, in hundreds of countries.

And the fact that it even holds together at all

is kind of a miracle.

Right, well it holds together because we have standards,

and we have protocols

that you mentioned. Exactly, protocol standards

for how the equipment talks to one another.

And increasingly, certificate authorities

that help bootstrap the secure, encrypted

and communication between end hosts.

So there are a few of these sort of centrally,

kind of agreed upon kinds of infrastructure,

but for the most part, each network runs itself.

And certainly, we've heard about some countries

that impose firewalls

that don't let certain kinds of traffic out,

or certain kinds of traffic in.

So there's no global body that is regulating that?

Not really because each country

really can have it's own laws and its own norms.

And so they can decide,

like the Great Firewall of China can decide,

they don't wanna let certain content be accessed

by the citizens inside that country.

So if a country decides they don't wanna answer a request

for a particular domain name, they say,

Hey, I don't want to let someone know the IP address

of this website.

They can decide not to let those answers be delivered

inside their country.

And so encryption plays a role in helping people

keep their privacy or prevent surveillance,

but it's not perfect.

It's often possible, still, to know a fair amount

about what people are doing, even if you can't look

inside the envelope at the letter that's written.

I mean, even you could just tell

that two people are communicating

even though the traffic itself is encrypted.

So you don't know what they're saying,

just even knowing two devices are communicating.

Exactly, and in fact, if you look at say,

the sizes of the transfers that they're doing,

you may know, hey, I'm talking to Netflix.

And by the way, this is the length of the movie I watched.

This is the size- So you can infer

or guess a lot of things.

Exactly.

You're one of the most awesome networking researchers

that I know.

I'm curious, just to pick your brain,

what do you think are some of the hot topics

in networking research?

Where do you think the field is heading?

Yeah, I'm excited about the convergence

of wireless communications, cellular networks, Wi-Fi

with networking and cloud computing.

And in particular, we're seeing in edge computing,

a convergence of all three.

Where you might have a mobile phone

or a drone or some other kind of device connecting

over the wireless medium directly to a network

that connects you directly to the server

that might run your application.

So you want the computation close

to where the endpoint is.

Exactly, and I think that what's now exciting about that

is all three of these technologies, wireless, networking

and cloud, which are normally three different communities,

three different sets of technologies,

three different sets of standards or practices,

now have to work together in close harmony

to be able to service applications that are really critical

and that that might be interacting with the physical world

in ways where safety is a potential concern.

You know, we've had cellular networks now

for 20, 30 years.

So when we hear about 5G,

what's trumpeted the most is the fact

that oh, super high bandwidth, right?

But I sense that the exciting things

are more than just the network being faster.

I agree.

It's both the high bandwidth, it's the low delay

so that you can have these applications

that interact with the physical world

and need answers in real-time.

It's about having the compute really close

so that you can integrate computation and communication.

It's about having more coverage.

Coming back again to the softwarization.

SDN and softwarization

is a maybe a little bit behind the covers,

that you wouldn't normally see it as a user going

from 3G to 4G to 5G.

You just see an increase in speed.

But yet, the way the network is now being managed again,

I think is bringing the cellular networking world

sort of into the internet world

in terms of the softwarization-

Completely agree.

I think the bringing in of compute and storage

is important too.

I think when you think just about networking,

it really is often just one part of the IT,

the information technology ecosystem.

Is there's often compute and storage as well.

And so, I think now there's an opportunity to have all

of those parts of the infrastructure work together

towards an even higher level goal.

And so I think it's a really exciting time

to be in the field 'cause now,

the plumbing is getting close to the application

in a way that it wasn't before.

[upbeat music]

So I really hope you've enjoyed this video,

and I hope you've also understood the internet

is part of the worldwide global communication fabric.

It's absolutely fascinating how it works.

[upbeat music]