linux/SMP/arla difficulties

[Magnus Ahltorp]

> o once arla is up and running, the system crashes after anywhere from a few
> minutes to a few hours - no obvious correlation with AFS activity. 

I can think of two things that may cause this to happen in such a situation: 
* Callbacks from the server
* A sudden need for the VFS to clean some dentries out.

> o before the machine finally hangs, tons of messages appear of the
> form "sending wakeup: ..." (which are apparently generated in xfs).

This is probably the callbacks or the dentry deletes, but it's hard to
tell without log messages.

> o during this same time, arla goes nonlinear and chews up all the CPU

Does arlad chew up all CPU for a short time, or does it do that infinitely?

> o when arlad is first started (using the startarla script), the
> sysname can't be changed successfully for at least a minute.

This may be caused by the precreation of cache nodes, but that should
be a background thing.

> Is there some additional info that I could gather that could help
> diagnose things?

The xfs logs are very much useful, especially when combined with the
arlad logs (--debug="all,-cleaner").