Newsgroups: comp.parallel.mpi
From: dwolff@baton.phys.lsu.edu (David Wolff)
Subject: ch_p4 on workstation clusters problem
Organization: /etc/organization
Date: 5 Dec 1997 20:13:11 GMT
Message-ID: <669n8n$14ce$1@its1.ocs.lsu.edu>

Hello,

  I'm encountering some strange behavior with MPI on a cluster of
DEC alpha workstations.  I'm working with a large code and 
occasionally the program will freeze.  I have tracked down the
problem to a particular MPI_Barrier call.  When I added a printf
statement previous to the MPI_Barrier call, the freezing problem 
went away.  Others who use this program have not seen this 
problem (they are running on large parallel machines, not 
clusters of workstations).  It is my hunch that there is 
something in the larger communication lag time that causes
this condition, and the extra time to execute the printf 
statement somehow equalizes that.

Does anyone have a suggestion on how to solve
this problem?  Has anyone seen this behavior before?

  David Wolff
  LSU Dept. of Physics