offtopic question about high density wifi

[david at lang.hm]

On Tue, 26 Jan 2010, C. Scott Ananian wrote:

> On Mon, Jan 25, 2010 at 10:31 PM,  <david at lang.hm> wrote:
>> I've read through these, and they have a lot of useful info. I do have
>> good RF experiance (and even some halfway decent tools for looking at
>> things), but I didn't know what, if any limits there were on the number of
>> clients other than what can be supported by the available airtime.
>
> Lots of very subtle protocol limits, having to do with all sorts of
> random mostly-timing related parameters where the 802.11 spec gives
> implementers a lot of freedom to choose arbitrary values.  *Plus* the
> details of all your clients.  There are all sorts of fancy algorithms
> you can use to tune your AP, but all it takes is one bad client and
> everything goes to hell.

right, this is the 'simple' issue of RF congestion

> And that's completely apart from the arbitrary software limits that
> some access point manufacturers include, in order to differentiate
> their "consumer" and "professional" product lines.

this is the issue I am most worried about. the RF issues I can measure 
(and to the extent they are outside my control, like the clients, I just 
have to live with), but finding that an access point only allows 64 
clients and then rejects everything after that is not something I know how 
to predict ahead of time.

are there any tools that can use a linux box (or access point running 
something like openwrt) that can let one piece of hardware simulate being 
many clients to find limits like this?

> So, basically: theory is no substitute for experience.  It's not
> really the protocol that's the limit, it's the particular choices that
> particular access point makes and the choices that "common" clients
> and "common" software make.  So your best bet is really to (a) find
> someone who's done it before, and slavishly copy their setup
> (variations that you think are trivial, like between firmware
> revisions, may in fact be critical), or (b) find a company who's
> invested the time and money to figure out all the variables and do the
> real world testing, and fork over the $$$ for the "commercial quality"
> or "pro grade" or whatever-they-call-it access point with a guarantee
> about the number of clients it can support.

part of the reason I was asked to step in on this is that the last time 
they ran their own wireless, they went out and purchased a bunch of 'pro 
grade' access points, and things did not work (I think I've discovered 
that they limit connections to 64 nodes)

last year one of these big name companies came in and did the wireless, 
but they didn't do a proper site survey (their people showed up on setup 
day with bags of equipment and started asking about what rooms would be 
used then), the results weren't very good.

part of the problem is that, on the RF side, things depend so much on your 
exact environment (will the hotel turn off their wireless, what is the 
impact of the wireless from the 5 hotels adjacent to you, how much do the 
walls of this particular building block/bounce the signal, etc). no 
professional is going to tell you how to set things up without doing a 
bunch of (expensive) testing.

useing what other people use in terms of equipment can help, if you have 
the budget to buy all new equipment


I very much don't expect to get it 'right' this year, but I am hoping to 
avoid outright collapse (well, at least beyond the predictable 'people are 
trying to use more bandwidth than you have' issues)

> And a couple of accessible wired switches and wired internet kiosks
> will go a long way toward mitigating your downside if it turns out
> that your wireless totally melts down under load.

those will be available where it's appropriate.

> I've attempted a couple of mystery hunts
> (http://en.wikipedia.org/wiki/Mystery_Hunt) with ~50 clients on one
> access point and can vouch that that's above the workable capacity of
> even consumer-grade access points.  (We had a fancy commercial grade
> access point this year with 10s of compact antennas and it did much
> better.)  As Chris noted, we did testing at OLPC and found that even
> 30 clients was pushing for most access points.  At the time, I backed
> that up with a literature search and could cite the various parts of
> the 802.11 collision-avoidance algorithm which melted down in the 10s
> of clients.  I can't cite chapter and verse any more, sadly.

the problem with the 'lots of directional antennas' approach is that you 
hit the point of diminishing returns because while the directional 
antennas make the access point only see a few of the clients, the clients 
are all seeing each other, and so you end up with the hidden node problem.

David Lang