Distributed Congestion Control

Loading...

Из курса от партнера Princeton University

Networks Illustrated: Principles without Calculus

39 оценки

Princeton University

Networks Illustrated: Principles without Calculus

39 оценки

What makes WiFi faster at home than at a coffee shop? How does Google order its search results from the trillions of webpages on the Internet? Why does Verizon charge $15 for every GB of data we use? Is it really true that we are connected in six social steps or less? These are just a few of the many intriguing questions we can ask about the social and technical networks that form integral parts of our daily lives. This course is about exploring the answers, using a language that anyone can understand. We will focus on fundamental principles like “sharing is hard”, “crowds are wise”, and “network of networks” that have guided the design and sustainability of today’s networks, and summarize the theories behind everything from the social connections we make on platforms like Facebook to the technology upon which these websites run. Unlike other networking courses, the mathematics included here are no more complicated than adding and multiplying numbers. While mathematical details are necessary to fully specify the algorithms and systems we investigate, they are not required to understand the main ideas. We use illustrations, analogies, and anecdotes about networks as pedagogical tools in lieu of detailed equations. Please note that per Princeton University policy, no certificates, credentials or reports are awarded in connection with this course.

Из урока

Controlling Congestion in the Internet

The Internet has many important tasks to manage, like routing packets (discussed in the last module) and controlling congestion. This workload is modularized into different functional layers, each responsible for performing a different set of functions, as we will see in this lesson. We will also look at the principles of congestion control, managed at the transport layer.

Divide And Conquer2:10

Layered Protocol Stack4:13

Transport & Network Layers4:42

Processing Layers2:17

Controlling Congestion4:03

Traffic Jam & Bucket Analogy3:02

Sliding Window2:36

Cautious Growth of Window Size6:51

Inferring Congestion5:46

Congestion Control Versions2:16

Loss-Based Congestion Inference5:19

Delay-Based Congestion Inference: Part I6:43

Delay-Based Congestion Inference: Part II5:33

Delay-Based Congestion Inference: Part III4:51

Distributed Congestion Control5:14

Познакомьтесь с преподавателями

Christopher Brinton
Lecturer
Electrical Engineering
Mung Chiang
Professor
Electrical Engineering

So, let's us consider zoomed out due over time of TCP Reno and TCP Vegas.

We've used these as the two examples just before we use Reno

for the lost base congestion difference and Vegas for the delay base.

so if we sort of zoomed out that picture out a little

if we kind of get the following for Reno and following for Vegas.

where as under Reno we would, you know, we would

increase it slowly and then we would

sharply decrease it once we get congestion.

And then we would gradually build it back up.

Sharply decrease it.

Gradually build it back up. Sharply decrease it.

Right. And we get sort of the zigzag pattern.

Now, obviously there's, you know, this is like a staircase up in here.

but this is just over time really zoomed out.

And then on the Vegas side of things of course, this is

still a staircase moving up but then eventually we would start to.

We would start to hit a centerpoint right where we would hold

constant for a little bit and then we would start to decrease.

So, it kind of smoothes this out a little bit to Vegas because we're responding

in a finer granularity, each acknowledgement we're

looking at we're either increasing or decreasing the.

Window size on the spot right there to send out again.

And so this is a typical property in one

of the typical comparisons we use between Reno and Vegas.

so this is really showing that Vegas can

smoothen out this trajectory and it's the advantage of

using a non-binary congestion signal, right because with

TCP Reno, we're just using one or zero, right?

It's either there is congestion or there is no congestion.

But with Vegas we're smoothening that out more because

we are also considering the magnitude of congestion, right?

is how late is this packet, right?

Is this packet early? Is it late?

So on, and then we can increase quicker or smaller accordingly.

And so but another thing to point out though is that it's

not always saying TCP Vegas is going to be better than TCP Reno.

it depends on the implementation, it depends on the parameters as well, right?

So, for instance, what do you set that threshold value to be, right?

but as long as you tweak and tune

the parameters properly to the right setting you would

tend to see that Vegas would have this smoothing effect.

And that it may eventually converge to an equilibrium

for each of the centers that are on the system.

So, that brings us to the next point of distributed congestion control.

and for this, let's think back to DPC, our distributive power control for a minute.

in that case, where our objective is to

match the measured to desired signal to interference ratios.

And it was

an iterative algorithm, DPC algorithm, right?

It was a sing, it was not like a single step algorithm.

we would keep, we would compute the measured SIRs.

Send them back to the transmitters. Then they recompute their powers.

Then we recompute the SIRs.

Then recompute the powers and. It will keep going back and forth.

And it was also distributed, right?

as it's name because it's executed on the device side.

And we know that we could

reach equilibrium amongst the transmitters, right?

And so, distributed congestion control, it's

really how we implement congestion control.

the objective is to match measure to desired transmission rates.

And it's really very similar to distributed power control.

they're doing something different but the idea is similar, right?

Is that we're trying to match transmission rates.

So, where the

[INAUDIBLE]

match transmission rates we're getting to what's we want to have ideally.

instead of, measuring matching SIRs, we're matching transmission rates.

it is an iterative algorithm. it is also an iterative algorithm.

We can make it iterative.

And we can also make it distributed, right?

So, we can make each, each sender is

doing their own congestion control at the TCP layer.

And we can also

reach an equilibrium.

And so it's it's similar situation that we saw in

chapter one or lecture one that we had a tower.

And if we had you know, cell phones then

we would be able to feed back and forth between

them and that they were all causing interference on one

another, which is in the form of a negative externality.

And here, it's just that we're viewing the internet and we

have devices, which are also, you know, that could be cell phones too.

But also like laptop computers.

can't draw to save my life but it's, you get the idea and they're all sending

and receiving feedback signals from the internet itself.

in terms of how much congestion there is to how to change, so.

They also don't have to communicate with each other and they, they don't

They just see kind of what the network is telling them.

And then the TCP layer uses that.

So, don't get confused.

This is DPC.

We also saw the the single set TPC algorythm, which led us to DPC and

TCP try distributed congestion control protocol not

to be confused with transit power control algorithm.

And so we don't have time to discuss the multiple transmitter implementation.

but actually there is something quite similar to

the way DVC works, an implementation we can

use over time, to adjust what the sending

rates would be in order to reach that equilibrium.

Ознакомьтесь с нашим каталогом

Присоединяйтесь бесплатно и получайте персонализированные рекомендации, обновления и предложения.

Coursera делает лучшее в мире образование доступным каждому, предлагая онлайн-курсы от ведущих университетов и организаций.

© Coursera Inc., 2018 Все права защищены.

Загрузить из App Store

Загрузить в Google Play