[MPICH] MPICH2 smpd phrase disappearing

[Raymond Chan]

Hi all,

 

I'm a newbie to the list (and MPICH2 for that matter), and hope someone can
help with a problem I've recently encountered and don't know why:

 

I used to have an MPICH2 (smpd) app (mpiblast) run phenomenally on my ROCKS
cluster.  I also got it integrated w/ Sun Grid Engine well, and it was
running great for a while.  Every user has a .smpd file in their home
directory containing a different passphrase with only RW access to that
particular user, as indicated in the MPICH2 documentation.  Since I am
having problems as of late, I now have two test users I've been using to try
to narrow things down with.

 

As I start running a few parallel jobs between the users, eventually a job
hangs-I found out that the passphrase inside that user's .smpd file
disappeared!  The file has of course, the nodes that it contacted for the
parallel job, but the passphrase that I explicitly put in a while ago is now
gone.  I add back a passphrase and the whole process works again.
Eventually, the passphrase of one of the users disappears again, and the
cycle starts over again.

 

Is there any reason in the MPICH2 implementation where a passphrase would
get deleted, or maybe a particular line gets cut out in the .smpd file that
would coincidentally always be my phrase?

 

Hope someone has an idea.  

 

Thank you all for your time,

-Ray C.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20060126/585c80f6/attachment.htm>