[Pw_forum] MPI & disk unavailability
Axel Kohlmeyer
akohlmey at vitae.cmm.upenn.edu
Mon Nov 28 20:46:22 CET 2005
KK> > locally. The "local" disk may be (and most often is) a networked
KK> > filesystem; but again, this is handled by the operating system, and
KK> > completely transparent to Espresso.
KK>
KK> OK. But here is my interpretation. Suppose cpu0 wants to write
KK> everything to the NFS disk, which is not there. Then, should not this
KK> write hang and wait till the disk become available? This is not really
for this to work, you first need to configure NFS for reliable
communication by using hard mounts, that do not timeout easily.
KK> far fetched, such a behaviour easily happens for serial executables.
KK>
KK> Now, if cpu0 indeed hangs, but other MPI cpus do not get any answers
KK> within the timeout frame, they start to quit because cpu0 is not
KK> responding.
this handled by the MPI _implementation_. you may want to try a
different package (there are several others besides MPICH which is
IMNSHO somewhat clunky , see e.g.:
http://www.lam-mpi.org/mpi/implementations/shortlist.php ),
or hack your local installation to increase the delay before
there is a timeout and the library considers the cpu0 process
as dead.
axel.
KK>
KK> Any ideas?
KK>
KK> Kostya
KK>
KK>
KK>
KK>
KK>
KK>
KK> __________________________________
KK> Yahoo! Music Unlimited
KK> Access over 1 million songs. Try it free.
KK> http://music.yahoo.com/unlimited/
KK> _______________________________________________
KK> Pw_forum mailing list
KK> Pw_forum at pwscf.org
KK> http://www.democritos.it/mailman/listinfo/pw_forum
KK>
--
=======================================================================
Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
Center for Molecular Modeling -- University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.
More information about the users
mailing list