Re: wi polling

On 2/28/2003 11:44 AM, John Polstra wrote:
> In article <[email protected]>, Lars Eggert  <[email protected]> wrote:
> 
>>Sorry for being unclear: By "livelock" I mean the CPU spends all its 
>>cycles on receive and hostap processing, to a point where userland 
>>processes are starved. The box doesn't panic and becomes responsive 
>>again when network load goes down.
> 
> Are you using IPSec on the interface?  I've noticed that's a
> reliable way to produce the symptoms you described.

Makes sense, IPsec increases the per-packet processing overhead
significantly.

I use IPsec on the box, but not on the interface: The box is an access
point/router, and the uplink is over an IPsec tunnel. However, the
livelock happens for local traffic between two wireless nodes. The
uplink capacity is too low to cause livelock.

Lars
-- 
Lars Eggert <[email protected]>           USC Information Sciences Institute