You pick up your iPhone while waiting in line at a coffee shop. You google a not-so-famous actor, get linked to a Wikipedia entry listing his recent movies and popular YouTube clips of several of them. You check out user reviews on Amazon and pick one, download that movie on BitTorrent or stream that in Netflix. But suddenly the WiFi logo on your phone is gone and you're on 3G. Video quality starts to degrade, but you don't know if it's the server getting crowded or the Internet is congested somewhere. In any case, it costs you $10 per Gigabyte, and you decide to stop watching the movie, and instead multitask between sending tweets and calling your friend on Skype, while songs stream from iCloud to your phone. You're happy with the call quality, but get a little irritated when you see there're no new followers on Twitter. You may wonder how they all kind of work, and why sometimes they don't. Take a look at the list of 20 questions below. Each question is selected not just for its relevance to our daily lives, but also for the core concepts in the field of networking illustrated by its answers. This course is about formulating and answering these 20 questions.
Air Interface Example
Loading...
En provenance du cours de Princeton University
Networks: Friends, Money, and Bytes
61 notes
À partir de la leçon
Why Am I Only Getting 3% of the Cellular Speed?
Advertised network speeds are typically only those that can be obtained at the physical layer under ideal channel conditions. In this lecture, we will study various factors that impact the actual speeds we obtain at the application layer under realistic channel conditions.
Rencontrer les enseignants
Mung ChiangProfessor
Electrical Engineering
Christopher BrintonLecturer
Electrical Engineering
Now remember our goal is to find out the answer to the question, why does WiFi
work so much slower at hot spot than at home in general.
So we will like to find out the performance of this S, okay, which we
just derived an explanation approximation expression of it,
as a function of the number of stations, number of users out there, N.
So now we're going to walk through a numerical example.
First we will define the length of the time slots.
Suppose we are using 802.11G WiFi, which transmit at 54 megabit per second.
And here the relevant timing parameter. Okay, each
a single time slot is nine microsecond. Everything is microsecond.
Okay? And then the SIFS is ten microsecond.
And the DIFS is six plus two time slots units which is 28 microseconds.
Okay so now we can look at TB, that is the time slot length for a back up slot
with idle channel. Okay that's the length of DIFs,
so that's just 28 microsecond. What about the time slot for a successful
transmission? That means we have the data frame self
plus waiting siffs plus acknowledgement packet plus waiting DIFs This totality is
the TS. So lets look at the time it takes to send
this data. Now the data consist of a header and then
the payload. The header, and then the payload.
Okay. First of all there is a 16 microsecond
physical layer preamble. You just waste sixteen microseconds as a
guard band at the top. And then there is a 40 bits physical
layer head, a physical head with information about a physical layer
configuration. The first 24 bits is sent at a much lower
speed instead of 64 is sent at six megabits per second so that it is has a
much higher probability of being decoded at the receiver end.
So there is 24 of them over six megabit per second, okay?
Since this microsecond, this is megabit per second the factors cancel each other
out and just write us 24 over six then there is sixteen remaining.
While plus 240 MAC layer, link layer had, okay?
Plus 32. Bit of error correction codes in the link
layer. All these are sent at the rate of 54
megabit per second. Plus of course, the actual pay load.
This part is L bits okay and we're going to later assume that L is
Basically, 8,192 bits, okay? So it's a typical value of the number of,
bits in the payload, okay? So now payload is sent at 54 megabit per
second. So this boils down to 50, 25.33 plus the
payload. Say, this number of some other number
over 54. Microsecond.
Okay. That's the length of TS.
So finally, what about TC, collision? [COUGH] Well there's actually a couple
different variations on how pe, people define TC, but I think that the proper
way is to define it as essentially the same as TS, because you have to wait
until the disappear is over before you can guarantee that the acknowledgement is
indeed not sent back to you. So we have defined these parameters.
Okay? Now let's also define some other
parameter. Let's say the maximum number of back off
stages you can have in exponential back off is three.
So you can multiple by two by two by two then you stop and declare the frame is
lost. The minimum window of time they need to
wait is, let's say two to the four minus one.
Okay? In our analysis, we ignore this minus one
factor but here we can just say incorporate that and that is fifteen.
And we just define error to be 8001 to 92 bits.
So now we have all the parameters, and therefore we can plug it in.
We can plug in the formula. First, we can plug in the formula of the
tau and C solution turns out we can solve numerically tau to be 0.0765.
That is basically a 7.65% as the contention probability, the probability
you'll transmit as a single station. Okay.
And then this lead to PTPS calculation, which leads to the S calculation,
together with all those constants. Okay.
The exact blown out form of the formula is in the textbook.
And, or you can just verify that through your own, calculation, okay?
So now, we're going to look at this S as a function of a few things.
First of all, as a function of N, the number of user stations,
or in other words the impact of crowd size.
So now, I'm plotting S in meg-, megabit per second over N here.
If you look at the aggregate, okay,
as, as a function of n for all, then it goes up, and then goes down.
Now goes up is actually better understandable, because I have got more
stations. But it quickly starts to go down.
This is the point where, basically the tragedy of commons kicks in so much that
adding more users will reduce even the total throughput across all users.
And this happen around eight users. And if you look at the total throughput
divided by M, okay,
as in over M. That's the average per station
throughput. Then, they actually always go down.
It never goes up. Why?
Because you add more user and more interference.
What is important is that it goes down so rapidly.
As you go from like two, three users down to ten fifteen users.
It went down from 25 so megabit per second now.
Notice not 54 because 54 is the physical layer of speed.
Okay. After the o, the overhead it goes down to
about 25 realistic speed. This is the theme we'll pick up in the
next lecture. Okay.
In today's lecture we'll notice the shape of this drop.
This drop is dropping very rapidly. Okay.
The, the point of going down all the way to only one or two megabit per second.
So no wonder in a busy hot spot. the average per station throughput is so
low, because despite all the smart ideas, the SMA random access controls strata
commons in a very inefficient way. Now few more charts for example we can
also measure S as a function of aggressiveness.
One way to look aggressiveness is to look at the minimum window size you have to
wait up to that point. Now we that for different size of the
crowd all happens that initially, okay, if you assuming W mean a bigger means you
maked it You make this this contention less
aggressive, then the flufoot actually goes up.
Okay? That's very good.
But at some point it will go down because it is so non-aggressive, you're actually
wasting idle resources in the network. So there's a point beyond which deemed
more polite that it actually hurts your through put.
Again, very typical of a cocktail. You don't want to be too aggressive but
if you're too non-aggressive then you're just wasting time slots.
And as the crowd gets bigger and bigger. Okay, we see that.
The. Range of W mean before it start to band
down becomes longer and longer. That means as the crowd gets long- bigger
and bigger it pays more and more to reduce aggressiveness.
Another way to look at aggressiveness is look at
The maximum number of back off stages that you allow is you make B bigger.
You tend to increase the average contention window size and therefore
become less aggressive and you'll see a similar behavior here.
Okay? as the crowd becomes bigger and bigger,
the impact is more prominent. Okay.
The Throughput actually becomes bigger as you
become less aggressive. Okay.
The impact of B however is less prominent than impact of W mean there.
Finally, as a function of the payload size L.
Okay. We were talking about somewhere around
here. Okay.
So get around 25 mega per second. Now you see a monotonic increase curve,
because more payload means less overhead, relatively speaking.
But this is a misleading chart, because remember all the way back early in the
lecture we did not model the actual interference or collision phenomena
accurately. As the.
Payload gets bigger and bigger actually, the chance of collision goes up.
Because the chance of two packages overlapping in time goes up as it takes
longer to transmit the payload. Okay.
If you incorporate that factor, this actually will start to bend over and
downward. So in summary, what we have seen is that
in wi-fi, interference management is done through random access rather than in
power control accelerator. And a big part of the reason is because
it's operating in the unlicensed spectrum.
There are a few very good ideas, including randomized and exponential
backoff. Including differentiated wait and listen
intervals. Including limited explicit message
passing. Which, by the way, the RTS/CTS is not
always enabled. And that may also explain some of the
inefficiency of throughout in hot spots. But, it's got a big limitation.
Now, we went through, Simple, relatively speaking.
But, still a little bit involved approximation of the throughput, and we
saw that this throughput per a station as a function of n drops rapidly the
performance decrease very fast as the contention intensifies.
Even, as we go up from several users, to just say, ten or fifteen users.
And, this is underlying reason why The performance in hot spot tend to be poor
unless you don't have a large crowd, it so happens.
And we see a fundamentally different way to do distributed coordination in taming
this tragedy of commons. So now we're going to wrap up our
wireless lectures with one more lecture on a very practical important question,
what is the actual speed that I can expect on my cellular, LTE or 3G network.
That will be the next lecture, I see you then .
Explorer notre catalogue
Rejoignez-nous gratuitement et obtenez des recommendations, des mises à jour et des offres personnalisées.
Coursera propose un accès universel à la meilleure formation au monde,
en partenariat avec des universités et des organisations du plus haut niveau, pour proposer des cours en ligne.
© 2018 Coursera Inc. Tous droits réservés.
