Dan,

The reason that PSP clients "grind to a halt" with long Beacon intervals
can be sussed out with a bit of effort.  Think of what happens with a
TCP connection.  The sender shoots out a bunch of packets, to fill the
window negotiated at the setup of the connection.  Then the sender has
to sit and wait for TCP ACKs to return from the receiver, before it can
send a single byte more.  

Put an AP between the sender and the PSP client that is the receiver and
that window-full of bytes sits in the AP until the PSP client wakes up
and pulls the packets out, one packet at a time.  Then the client's
protocol stack processes the packets and sends back the TCP ACKs.
Typically, the client sees nothing more to send and goes back to sleep
until the next time the AP tells it there are packets buffered for it
(by setting the bit in the TIM enumerated by the client's association ID
(AID) in the next Beacon).  This is the best case.  If the sender is a
bit late getting its packets out and misses the transmission of the
Beacon, those packets will sit fir a full Beacon interval, waiting to be
announced in the following Beacon.

So, the soonest the client can receive another packet from the sender is
right after the next Beacon.  With a typical window size of 8 kbytes,
here are some maximum TCP throughputs for various Beacon intervals.

Beacon interval   Throughput
100 kus            80 kbytes/sec (10 window-fulls per second)
200 kus            40 kbytes/sec
500 kus            16 kbytes/sec
1000 kus            8 kbytes/sec (can anyone say 56k modem ;^)

I hope this has been helpful.

 -Bob

Bob O'Hara
Black Storm Networks
110 Nortech Parkway
San Jose, CA 95134-2307

Phone:  +1 408 941 0500
Mobile: +1 408 218 4025
Fax:    +1 810 277 4718
email:  [EMAIL PROTECTED]
 
 

-----Original Message-----
From: Dan Lanciani [mailto:ddl-bawug@;danlan.com] 
Sent: Sunday, October 27, 2002 12:19 PM
To: [EMAIL PROTECTED]
Cc: [EMAIL PROTECTED]
Subject: Re: [BAWUG] FreeBSD box as Acess Point without SSID broadcast


|While Russell is correct, there are reasons to not beacon SSID.

One possibly good reason to disable broadcast SSID is simply to avoid
confusing
clients that you do not want to connect (and that have no desire to
connect to
you).  My clients started acting odd when a neighboring motel set up
some
public access points in spite of the fact that my clients were set up to
use
only my own SSID.  It turned out that they were getting hung up timing
out on
association attempts with these public access points (and counting up
SSID
mismatches) when they should have been talking to my access point.
Clearly
it's pretty stupid of the firmware to keep trying the wrong SSID on an
access
point that is willing to tell its SSID, but that's what they did.  The
resulting
delay caused the client in my car to fail to associate with my own
access point
before I turned into the driveway, confusing my home automation system.
I
ended up restricting the clients to my specific access points which is
kind of
a pain since MAC addresses have to be everywhere...

|That is,
|to keep junk out of the air.

I'm not sure it accomplishes this one.  As far as I can tell, most
access
points that allow you to disable "broadcast SSID" mean that they will
not
accept the null (= broadcast) SSID from clients in probes.  The setting
does
not change the rate of packets generated at idle.  Having just ordered
one
of the sniffer boxes advertised here recently, maybe I'll be able to see
for
certain soon. :)  (For some access points--like the ones mentioned
above--
allowing broadcast SSID seems to make them accept any SSID rather than
just
the null and configured ones.  At least in some contexts.  Very
confusing...)

|In a congested environment (like here in
|Cambridge) this can make a difference. At the least, set your beacon
|interval to something reasonable (2 per second is good).

I've experimented with long beacon intervals and they kill PSP clients.
I
don't quite understand what is going on yet, but the clients just sort
of
grind to a halt if the beacon interval is significantly longer than the
default 100Kusec on my Aironet access point.

|I think a patch to allow disabling the beacon would be a good idea.

How do you convince the clients that the lack of beacons doesn't
indicate
that the access point is dead?  The active beacon seems to be on of
those
unfortunate things which wasn't absolutely necessary for DSSS in the
first
place (active probes are fine, no?) but which has come to be depended on
for a
variety of functionality.  Some of that could be changed, but PSP's
dependence
is pretty deep...

                                Dan Lanciani
                                ddl@danlan.*com
--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless

--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless

Reply via email to