Message: 1
Date: Thu, 3 May 2007 10:18:13 -0400
From: "kati london" <[EMAIL PROTECTED]>
Subject: [nycwireless] throughput & range question
To: <nycwireless@lists.nycwireless.net>
Message-ID: <[EMAIL PROTECTED]>
Content-Type: text/plain; charset="us-ascii"
Hi,
I'm trying to expand on a current hardware platform which includes:
Network card: EMP-8602 6G: 802.11a/b/g 400mW High Power mini PCI Card
Antenna: 1 x 8 dBi Fiberglass Omni N Female
Board: Soekris Net4801
Here are the issues:
I am interested in maximizing range over a flat, concentric circle area
- I want the biggest radius possible. I have a relatively large space to
cover.
Get bigger antennas.
I want to maximize throughput -- in internal tests, I've seen about
2.2MB/s cumulatively, testing with a single client downloading a large
binary file and with several clients downloading a large binary
simultaneously. This is the crux of the issue -- wifi usage is generally
about several clients downloading small files at different times, so you
can support relatively large numbers of connected clients without
throughput bogging down. Since usecase is specifically about several
people downloading large files at the same time, it's very affected by
WiFi's theoretical and practical throughput limits.
The source of the WiFi traffic is not important. One big file or
many little files makes little difference. They are all broken
down to packets. You can twick MTU ( Maxim Transfer Unit )?
Transport method will make a difference ( TCP or UDP ). FTP uses
TCP and SCP will use UDP. UDP traffic will be faster.
There's a chance I'm CPU limited on the Soekris board with several
clients connected at once -- this might be affecting our practical
throughput limits.
Don't think so? You can use my monitoring program to see what kind of
load you are putting on your Soekris:
http://64.124.13.3/projects/pfss/
It will show you a real time plot of your CPU and network load.
Install pfssh on the Soekris, pfssg on your monitoring PC, then run it
with 'ssh [EMAIL PROTECTED] pfssh | pfssg'. You will need gnuplot
installed on your local PC for pfssg.
I want to deal with the 802.11b/g situation: I'd like to mitigate the
effect of a b client connecting to the network. So far the proposed
solution is two wifi cards per access point: one pinned at G (which b
clients won't see), the other supporting B and G.
You need to be careful here. 802.11B and 802.11G use the same frequencies.
So the two radios will interact with each other. If you are using them
in a 'concentric circle' pattern, you will need to separate them both
psychically and by channel. Vertical is the best psychically separation,
say may be 15 feet. Use good coax, LMR400 or better, in short runs.
Channel separation will be determined by your site survey. Some people
say 4 channel separation is required? I use a 3 channel difference, but
I am also using directional antennas.
You may not get any better transfer rate using 'G' mode.
Any suggestions to deal with the above, additional gotchas we may not
have thought of, etc.
A network topology would be helpful. How many APs are you using? How
are they connected? What kind of traffic ( http, ftp, etc. ) do you
expect?
Kati
--
William Estrada
[EMAIL PROTECTED]
Mt-Umunhum-Wireless.net ( 64.124.13.3 )
--
NYCwireless - http://www.nycwireless.net/
Un/Subscribe: http://lists.nycwireless.net/mailman/listinfo/nycwireless/
Archives: http://lists.nycwireless.net/pipermail/nycwireless/
