802.11n Impact, Part 7 of 8 - Wireless Network Scalability
Now let's look at the network impact. I already alluded to the fact that 802.11n is much more spiky and erratic. This graphic shows you, in fact, from a real-world example what sort of coverage pattern you can expect. The dot over here -- the black over here -- shows you the location of the access point, and what you'll see, is if you're using ABG the rate-range curve is essentially more monotonic. The rate monotonic decreases as you go further away. But with 11n, because of the MIMO recombination, you might get low rates and then suddenly your rate might spike up, then again you get low rates, suddenly it might spike up again. The solution of predictability being much harder.
Now what is the impact of not being able to predict? You can't really put access points on different channels and assume that you can get the coverage pattern right because you don't know how far signals propagate. You don't know how signals interfere. The network itself has to have the ability, at packet-level timescales, to resolve co-channel interference. You cannot design it away by trying to put neighboring APs on different channels because you don't know who the neighbors are. The definition of neighbors are where signals collide. You don't know who your neighbors are, therefore you don't know who to avoid.
So now let's move onto the next key point, which again, it's an implication of the burstiness. Notice obviously there's a price to pay for every technique of advancement. 11n increases efficiency through burstiness, and therefore there are some tradeoffs. Now the graphic on the top shows you how you can get fairness with 802.11abg. Now, with a technical legacy wireless LAN network, if I'm an access point, I'm transmitting to three stations, I can transmit one frame to each. One, one, one. Round robin. With 11n, if you do round robin, you lose the benefit of bursting because one of the main principals of bursting is you only burst to one receiver. It is not that you just access the channel and you burst on the entire channel for a single contention round. Now, there is a WMN mechanism called TXR that allows you do that, but the core 11n way of increasing efficiency is by bursting a whole sequence of frames to the same receiver.
Now if you look at the graphic below, notice that even though you increase the total amount of frames sent, if you look in short timescales, you're essentially giving the channel to one receiver, bursting its queue out and then going to the next one and then going to the next one. So in other words, rather than it being a fast round robin, it is a slow round robin with large bursts. You will look at this and say that doesn't really matter because over a period of time, fairness is fine, so it equalizes. That's true, except that some protocols like TCP are self-clocked. The more you send, the more it will try to send back.
So notice that when you combine MAC-level burstiness with an end-to-end protocol that tries to eat as much as it can, notice there are huge implications wherein it is possible with a system designed wrong that those devices that do well keep doing even better, and those that don't do very well will get starved. So this is a really important implication. The second one is really a throwback to what happened in 802.11bg. If you have a B device and a G device, the B device is going to slow the G device down. Very similar, in 11n, you have your A or G devices working with an 11n device, they're going to slow the whole network down. Now of course, there is a very good way to solve this problem -- we've already addressed this in companion videos -- which is to try to do airtime fairness. Every device gets its own amount of airtime, and what it does with its airtime is its problem.
And then there are some more sophisticated things that vendors and driver manufactures focus on, which is the size of this bursting, both on the sender size and the receiver side which shows as the A-MPDU aggregates and the A-MPDU window size. It just tells you how much you can burst. And then, of course, finally, their adaptation. So it turns out there's about 50 times more combinations to test with 802.11n than there are with ABG. So now that we've gone through the basic technology and the basic network concepts, let's go back to the overall principle which is how do you design this network?
......Please See Interactive Transcript For Full Text......
For more info visit: http://www.merunetworks.com
Loading...
9:13
Contention Management Schemes: Part 1 - Single / Multiple APby merunetworks1,196 views
Link to this comment:
All Comments (0)