Re: Rx overruns on vortex 3com card ...possible causes??

On 23 Aug 2003, Madhav Diwan wrote:

> but this problem occurs on whichever unit is primary ..

What do you mean by "unit" ? A network interface ? If so, the above would 
translate in "whichever interface is eth0" ?

> I dont see any thing that particularly sticks out other than the shared
> inturrupt .

Well, 3 or 4 types of network chips in the same computer is not exactly 
common - they would each use a different driver. But I agree that it 
should work...

> but should that not cause overruns on all three interfaces
> that are using it.. not just the one? 

The overruns are caused by the card not being able to give away the 
packets to the computer fast enough. This heavily depends on the traffic 
on the cards, so if eth0 is much more loaded than the others that share 
the same interrupt, it's much more probable to have overruns on eth0 than 
on the others. Depending on the card/driver features, overruns might be
harder to hit, for example by a bigger size of the Rx buffers ring.
So comparing the interfaces in this way has no meaning.

One possible solution for your case is to recompile the network driver 
with a higher value for RX_RING_SIZE and eventually to increase the value 
of the "max_interrupt_work" module parameter. But this is just a bandaid, 
saying "I give eth0 bigger chances to survive a burst of packets, while 
possibly lowering the chances for other interfaces".

-- 
Bogdan Costescu

IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: [email protected]


_______________________________________________
vortex mailing list
[email protected]
http://www.scyld.com/mailman/listinfo/vortex