Wifi throughput in F10 and variation with AP and chipset

[Mike Cloaked]

I have recently been assessing the difference between several laptops, with
different wireless chipsets, in terms of wireless transfer throughput rates,
and I wonder if anyone can explain the rate differences I have been seeing,
and what they depend on.

Up till a few days ago I was using a Linksys WAG54GL Access Point running
WPA2 encryption, and with this setup I was connecting from 4 different
laptops all running fully up to date F10 via NetworkManager in the Gnome
desktop.  On a file transfer using rsync from a single wired machine on the
LAN running up to date F10 and which has a 10/100 ethernet nic I was seeing
the following sustained transfer rates transferring a 400MiB file:
Dell M4300 with a 4965 wireless chipset 2.6 MiB/s
Dell D610 with ipw2200 getting at best 0.9 MiB/s
Samsung Q35 with ipw3945 chipset 1.5MiB/s
Samsung NC10 using the ath5k in kernel driver 1.7MiB/s
These were done with the AP set to "G only" rather than mixed mode.

Recently I installed a Linksys WRT610N Access Point, and the achieved
throughput went up dramatically, but also two of the machines would not
connect to the wideband setup for the main connection - the main connection
was set to wideband (wireless draft-N) using WPA2/AES, and this AP also has
a simultaneous guest connection that was running narrowband also with
WPA2/AES. The AP was set to mixed mode.
The Dell M4300 with 4965 now achieved 7.9MiB/s from 2.6 previously.
The d610 with ipw2200 would not connect to the wideband connection but only
to the secondary narrow band "guest" connection but was happy with WPA2/AES
and achieved 2.4MiB/s which is a considerable improvement over the original
speed (0.9) with the WAG54GL
The Samsung Q35 with 3945 now achieves 3.7MiB/s which up from 1.5
originally.
and the Samsung NC10 using ath5k achieved around the same speed as before at
about 1.7MiB/s but this can only connect to the narrowband AP connection and
cannot even see the wideband signal in the list in NM.

So it seems that achieved throughput is very dependent on which wireless
chipset is in the laptop, but also depends on which Access Point is used
also.  All the above were with the laptop only a few feet from the Access
Point in the same position with close to 100% signal level as indicated on
the NM icon graphic.

The throughput was measure both using the netspeed applet, as well as the
--progress option in rsync, and confirmed with the total data transferred
over the time taken for the whole file.

Can anyone suggest if there is anything that can be done to get the two
machines that could not connect to the wideband signal to be persuaded to do
so?  Or is this dependent on the current state of the drivers in those
cases?  Or is this hardware dependent?  I have been playing with various
combinations of wireless for a quite some time since well before F8 but I am
not an expert on the details of how the drivers in the Fedora 10 system work
or how closely they match the draft-N wireless spec?

Can anyone point me in a direction towards some information or answers to my
quest to understand/explain the above differences in performance?
-- 
View this message in context: http://www.nabble.com/Wifi-throughput-in-F10-and-variation-with-AP-and-chipset-tp22189940p22189940.html
Sent from the Fedora Test List mailing list archive at Nabble.com.