Re: [Madwifi-users] Performance

[Sam L. <sa...@er...>]

> Quoting Sam Leffler <sa...@er...>:
>
>> ...
>
>> FWIW I see >5 Mb/s running netperf under FreeBSD with the BSD version of
>> this code.  There are numerous performance anomalies with the code under
>> Linux that I've had no luck tracking down.  Many of these are mentioned
>> in the README file.  One important reason for making the driver available
>> in its present state was that I'm too overloaded to follow through on
>> many of these issues.  Other folks are going to have to step up and
>> contribute.
>
> Sam, are you saying you get only 1/10 the rated speed even under BSD
> (5Mbps vs 54Mbps)?  I was getting about 7Mbps on Linux in my tests...
> Windows of course gets in the neighborhood of 20Mbps+.

No, I was talking about performance over 11b.  For 11b I'm seeing close to 
full channel bandwidth with the BSD version of this code (which is very 
close to the Linux code and of course shares the HAL unchanged).  Basic 
performance results under BSD using netperf to talk to a station one 
wired-hop from an AP are:

5210/11a	23 Mb/s
5211/11a	15-16 Mb/s
5212/11a/11g 15-16 Mb/s

5211/5212/11b	5+ Mb/s

The 11a and 11g AP's were Atheros engineering prototypes.  The 11b AP was a 
Soekris 4511 with a Senao (Prism) card in it running FreeBSD in hostap 
mode.  The 5211 and 5212 performance results are suboptimal; the hardware 
is known to be capable of 23-25 Mb/s as demonstrated with Atheros' NDIS 
driver.  We are aware of this and are working on it.

Linux users should be able to get as good or better performance to BSD. 
Any problems that exist in the current distribution are in the driver 
and/or wlan code so there's nothing stopping someone from fixing the 
problems.  As I said before I'm simply too busy to work on this stuff right 
now--I hardly have time to answer email.  Any time I have will likely be 
spent working on HAL-related issues (like why folks are unable to use 11b 
in certain configurations).

	Sam