On Fri, 16 Dec 2005 13:04:21 -0600, Mike Hord wrote:

>...<
> Some people (including those who would claim to be IT
> professionals and should know better) are suggeting that
> a wireless network is adequate for our needs- namely,
> 120 students in one room at one time, accessing network
> resources.  They argue against our managing the wireless
> infrastructure, as well, saying that having the "experts"
> from central campus IT install and maintain "enterprise
> level" WAPs in each classroom will make the system
> somehow better than our doing it ourselves.

And what do these "experts" think is different about an "enterprise" WAP versus an off-the-shelf one?  802.11x 
is a standard, and it has limitations, and those don't extend to sending 6MB to 120 users at once!  Ya canna' 
change the laws o' physics, captain!  Even by calling it something that has a better marketing sound to it  
:-)  If they are on a single channel, only one WAP or user can be transmitting at any one moment in time, and 
if more than one tries, all will fail and will go into a wait-and-retry loop, so reducing the theoretical 
throughput dramatically.  They are going back to the days of 10Mb/s hubs on wired Ethernet, rather than the 
switches we use nowadays.

> The first real stress test occured yesterday.  In a room
> set up by the "experts" with "enterprise level" equipment,
> 120 students turned on their computers and simultaneously
> attempted to load a 6 MB slideshow from a shared drive.
> It was pandelerium.  I was on hand to provide assistance,
> and spent ten minutes running from one person to the next,
> advising patience, until some people's file finished loading
> and the network's bandwidth slowly thawed out and let some
> data through.

Ah, "Pandelirium" - my "you learn something new every day" item for today :-)  Any of these "experts" should 
have known that would happen - you have a half-duplex channel (send or receive at any one time, not both), 
which causes a geometric progression of collisions as traffic increases.  It's not rocket science, it's basic 
comms which has always been that way with CSMA/CD ever since it was first thought of.  Ask any proponent of 
Token Ring (or even ARCnet) and they will tell you why it's a problem.  :-)
 
> The bottom line, what I can't get people to understand, is
> that only so many WAPs can be in one place without
> interfering with one another

Yes, in the USA it's three, since that is the number of non-overlapping channels (In Europe & Japan it's more 
like 4 - I have no idea why the US chopped off the top few channels).  So the best case you could hope for is 
3 x 40 users, so in an ideal situation each user would get 1/40th of 54Mb/s, which a rough calculation shows 
would mean a bit over a minute and a half to get 6MB to everyone.  But of course the collisions problem makes 
this far from ideal, and I think you could reckon on at least 5 minutes in a real situation - maybe much 
longer.

> and that interference brings
> the bandwidth available to a particular WAP down.  For
> some reason, the general understanding is that these
> "enterprise level" WAPs can have multiple devices at the
> same frequency without causing a problem.

It may mean they can coexist, but it doesn't mean they can transmit (or receive transmissions for them) at the 
same time - just that they will step aside when the other one is transmitting.  When they are in the same 
room, it means that you effectively only have one of them because the diversity that would come if they were 
at opposite ends of the site is lost - every user machine will receive from both of them.

>  I've tried
> explaining it.  I've drawn diagrams, and explained
> signal-to-noise ratios and collisions and bandwidth
> sharing, and no one listens.  They just nod politely, and
> then at the next meeting, they dig in and say "we need
> to get the experts out here".

If they get real experts, rather than equipment salesmen or bluffers, then they will tell them what you are 
trying to say, because that is the real situation.

> The classroom in question has four WAPs, which means
> overlap out of the gate, but I'm not sure how bad it is.
> Best case, that gives us 216 Mbps to spread out among
> 120 students.  Factor in overhead and the worsened S/N
> ratio caused by overlap in the frequency bands, and I
> doubt we're doing that well. 

Nowhere near that well - even if there was no interference between WAPs, adding in the overhead for 
acknowledging packets would reduce it by at least 10%, and dealing with collisions for an individual WAP would 
reduce it much more.

> Now we have a prof who, for
> next semester, wants to use a remote desktop program
> to control the content on every screen of all 120 students.
> I think he's being over optimistic.  In a big way.

Ask him to do the calculations of the bandwidth he needs, and see if he will realise how feasible it is...

Cheers,


Howard Winter
St.Albans, England


-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist