[MLB-WIRELESS] Multiple APs

[Roger Venning]

Donovan, see the short paper I just found:
http://cwc.ucsd.edu/~rgholmie/WirelessPaper/WirelessPaper.html, paying 
particular attention to the section under 802.11 :-). This explains the 
way the medium access control works. I've also commented inline below.

Roger.

Donovan Baarda wrote:
<snip> 
> Yeah, that's what I thought. With all the AP's on the same channel, they
> would be colliding with each other all the time. Wouldn't you ideally want
> any two APs on the same channel not only out of each others range, but
> sufficiently out of range that no client in the middle could see both?
> 
Not necessarily true. If on different ESSID I believe that they use 
different spreading code, and the result is dimished throughput not as a 
result of an ethernet like 'collision' but rather because of greater 
'interference' ie. high detected noise level.

> This has probably been beaten to death on this list, but I can't find exactly
> what the benefit of AP's is. There is a lot of waffly stuff about how
> "infrustructure mode" relays all traffic through the AP to clients and how
> they can do bridging, but I still don't get it.
> 
They implement the Point Control Function which allows for higher 
channel utilisation due to reducing and coordinating the period where 
contention occurs.

> With all the clients in peer-to-peer mode, the whole system would be just
> like ethernet, with colliding clients backing off and retrying. This gives
> you effective bandwidth of 1/2 the total available bandwidth. This I
> understand.
Ahh... I don't believe it converges to 1/2, I thought the classic ALOHA 
result was something like 37% (with some relationship to e?) and that 
slotted ALOHA was twice as efficient or something.

> 
> With an AP in the mix, all traffic must be relayed through the AP. This
> means packets between clients must be transmitted twice, once to the AP, and
> again to the destination client. Now unless AP's negotiate some sort of
> collision avoidance, you still have 1/2 total available bandwidth from
> collisions, and then you halve it again because of the double transmits,
> giving you only 1/4 total available bandwidth. The only use I can see for an
> AP is relaying between two clients that are in range of the AP but not each
> other.
> 
I'm not certain that the A/P will *always* relay the frame. Also 
remember though, that in a typical enterprise situation that this isn't 
a really bad thing if it does becuase most traffic is destined somewhere 
else and not locally anyway.

> Is there some other voodoo that AP's do that buys you something that I've
> missed? Is it something to do with the incomplete direct conectivity between
> all clients? Surely this could be better handled by smart dynamic routing
> between clients in peer-to-peer mode?
> 
I would say yes (and am working on it as described at 
http://users.bigpond.com/r.venning/wireless.html), but there may be 
performance issues that need to be resolved as Ferni found out.

Roger.





To unsubscribe: send mail to majordomo at wireless.org.au
with "unsubscribe melbwireless" in the body of the message