[vortex] eth0: memory shortage

[Steven Timm]

>
> > In previous versions of the OS we also saw
> > eth0: too much work in interrupt
>
> This usually means that the 3c59x interrupt code is not called as
> often as it should, probably because interrupts generated by other
> devices are processed.
>
> > but now we are seeing them simultaneously across all
> > 240 nodes of a cluster.
>
> This is very strange... Do you have some application that uses
> multicast ? This could explain lots of packets to lots of nodes
> simultaneously...

We are using ganglia, as I said before there seems to be the most
activity on this when nodes enter and exit, but there is a
lot of nfs activity on shutdown and reboot as well.

Steve


>
> > eth0:memory shortage
>
> Hmm, I've tried to convince somebody on the RHEL mailing list that
> this had nothing to do with the driver and everything to do with the
> memory handling in general. The system is just short on memory and
> memory allocation by the driver fails. The recent RHEL3 kernels have
> had quite bad memory handling behaviour and this might be related.
> Supposedly the RHEL3U4 kernel that should come out very soon finally
> fixes memory handling.
>
> > Any difference in multicast handling?
>
> Not at the driver level at least...
>
> --
> 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: Bogdan.Costescu at IWR.Uni-Heidelberg.De
>
>