[OpenAFS-devel] AFS goes to WAIT on alpha_dux40

Harald Barth haba@pdc.kth.se
Fri, 09 Jan 2004 19:11:53 +0100 (MET)


> uname -a = OSF1 trp V4.0 1229 alpha
> sizer -v = Digital UNIX V4.0F  (Rev. 1229); Mon Jan  5 09:15:32 EET 2004

V4.0F is not the freshest. We have mostly V5.0A. OpenAFS 1.2.9 and
1.2.10 work nicely on V5.0A (for example for webb and email). However,
we do not use XDS.

I'm not that fluent with Alpha models: Multi or single CPU?

> After the transition to OpenAFS, it seems even worse.  The machine still
> responds to ping and the NFS filesystem it serves is still accessible

Do you really need to serve NFS from your AFS client WS? I'd rather not
and eliminate one more possibe problem.

> The application in question is XDS
> (http://www.mpimf-heidelberg.mpg.de/~kabsch/xds/).
> It causes no problems on x86 Linux or SGI, and we haven't seen this kind
> of AFS problem with the XP1000 in any other circumstances than those
> involving the running of this program.

You can debug this with syscall traces and the like. I'm curious, but
this seems extremely hairy. Do you need the 64 addr space or can you
replace the Alpha with a PC? That might be more efficient.

Good luck.

Harald.