Is valgrind supported on your system? If so, I suggest running your application using it with 2 processes in the job. I would start with MPICH. If possible, you might try MPICH2 as well. If valgrind doesn't report problems when running your application with MPICH or MPICH2, then try the same thing with MPICH-G2.
--brian |-----Original Message----- |From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf |Of leonid glimcher |Sent: Monday, September 24, 2007 11:05 |To: gt-user |Subject: [gt-user] mpich-g2/globus4.0.4 error??? | |Hi, | |i've been struggling to get a parallel MPI program (works with MPICH |1.2.7) to work with MPICH-G2 (same version). The new MPICH-g2 program |works on a single node, but not on multiple nodes (i've tried 2, 4, 8). |The program fails during an "MPI_Isend" call, which points to a bug |inside MPICH-g2, but what is interesting is that when you look at the |"gram_job_mgr_*.log" file on line 77 it says: | |*********************************************************************** |JMI: poll_fast: returning -1 = GLOBUS_FAILURE (try Perl scripts) |*********************************************************************** | |Does anyone have any ideas on what this means, or what kindof problems |this could be indicative of? i've been getting this error for a week and |i've tried attacking it from many different directions, but now i'm lost |and seem to be running out of ideas ... | |thanks in advance, | |~leo
