[Pvfs2-developers] mpi-io tests

Murali Vilayannur vilayann at mcs.anl.gov
Sun Sep 3 17:40:26 EDT 2006


Hey guys,
Has anybody observed server hangs with HEAD?
Even a killall -15 is not enough to shut the servers down :(
because the new termination code does not cause server to exit until all
the state machines terminate.

An easily reproducible test case (atleast on chiba) is to run an mpi job
with sufficiently large number of processes (say 5 or more)
where each process uses the vfs interface to read/write the same file.

I have checked in a program called io.c and iox.c to the
posix-extensions-branch that can verify the
hangs. It does require more than 1 server I think...

mpiexec -n <> io -f /mnt/pvfs2/File -b 1048576 -i 10 -u
mpiexec -n <> iox -f /mnt/pvfs2/File -b 1048576 -s 1 -m 0

Any clues appreciated..:)
Murali
-------------- next part --------------
A non-text attachment was scrubbed...
Name: io.c
Type: text/x-csrc
Size: 10874 bytes
Desc: 
Url : http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20060903/018f78d6/io-0001.bin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: iox.c
Type: text/x-csrc
Size: 18225 bytes
Desc: 
Url : http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20060903/018f78d6/iox-0001.bin


More information about the Pvfs2-developers mailing list