Do you have anything blocked on IO? While I was working on the infiniband stuff, I often ran into problems where things could get blocked on IO/network-traffic and cause servers to spin (polling for new data which never comes). There were also some issues that I came across where the kernel module would hang because an MD server was hung/spinning/dead. Usually restarting the server processes cleaned this up on the client-side.
If you cannot kill with a -9, IMO: you are blocked on IO or inside kernel land which implies a bug at some level. Im not sure if you've described it in detail already - I havent been in #pvfs2 for a while - but can you describe your setup in gross verbosity? Also, untarring on-the-fly on pvfs2 can be pretty brutal, and I seem to recall we highly recommended not running binaries from the pvfs2 filesystem - not sure if that has changed ? ~Kyle Kyle Schochenmaier On Tue, Jul 28, 2009 at 11:08 AM, Emmanuel Florac<[email protected]> wrote: > Le Tue, 28 Jul 2009 08:58:09 -0700 > Jim Kusznir <[email protected]> écrivait: > >> How do I fix this problem without replacing pvfs2? > > How do you access the cluster? are you using the kernel module, the > FUSE module, or else? Do you have any RAM usage problems? > > -- > ---------------------------------------- > Emmanuel Florac | Intellique > ---------------------------------------- > > > _______________________________________________ > Pvfs2-users mailing list > [email protected] > http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users > _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
