Re: [Madwifi-devel] WEP performance

[Stephane L. <ste...@co...>]

Th WEP encryption is actually performed by the hardware.  I cannot think of 
any reason why the ieee80211_wep_crypt() function exists.  It doesn't seem to 
be called in normal operations.

In any case, the performance hit is due to how the IV is set in 
driver/if_ath.c:ath_tx_start().  It calls get_random_bytes() for each IV, and 
on my system, that call takes 100ms.

With the attached patch, performance is back to normal.  get_random_bytes() is 
called only for the first IV, which is incremented thereafter.

Stephane

Peter Antypas wrote:
> Hi all.
>  
> This may not be the first time that you've heard this. It's about WEP 
> performance with this driver.
>  
> When in (b) mode, performance drops to less than 1/4 of the unencrypted 
> mode. Using a LinkSys 802.11b AP, throughput goes from 525 KB/s to 96 
> KB/s with no other variable other than turning WEP on/off. In (g) mode, 
> using a DLink 624 AP,  I go from 1.2 MB/s to 740 KB/s which is almost 
> half. In both cases, the stationary computer is WIRED to the AP to leave 
> the band open for the wireless node.
>  
> I have looked into the WEP encryption code in if_ieee80211subr.c and I 
> don't see anything that strikes me as terribly inefficient. I have 
> compared it to the linux-wlan code and it looks similar enough, save for 
> the fact that they use a statically defined CRC table, while the madwifi 
> driver initializes it (once, I hope).
>  
> So what could be causing this? I don't have a lot of experience in 
> kernel land, but I would be willing to dig my hands into it if you can 
> provide some pointers. Anyone care to comment? Sam?
>  
> Peter
>  
> Hi-Fi through Wi-Fi
> ===============
> www.joyfaktory.com <http://www.joyfaktory.com>
>  
>