[mvapich-discuss] vapi error on scyld configuration

[Abhinav Vishnu]

Hi Matt,

Thanks for using MVAPICH and reporting the problem to us. I believe that
you are using a *very old* version of MVAPICH with VAPI drivers. May i
suggest you to upgrade the MPI stack and the userspace libraries. We
have recently released MVAPICH-0.9.9 beta and planning to release the
final version in a week or so. Please download it from the MVAPICHwebpage.

http://nowlab.cse.ohio-state.edu/projects/mpi-iba/

Also, please upgrade the userspace libraries to OFED from the following URL:

http://www.openfabrics.org/downloads.htm

You may also need to upgrade your firmware for adapters and/or switches:

http://www.mellanox.com/support/firmware_download.php

Please re-run your applications with the settings above and let us know
the outcome of your experimentation.

Thanks,

:- Abhinav

* On Apr,1 Matt Funk<matze999 at gmx.net> wrote :
> Hi,
> 
> I am running a straightforward Finite Difference code on a machine with 
> infiniband. The Finite Difference code is such that i do the same 
> computations over and over again. I run the code on 32 procs with infiniband 
> on a scyld configuration machine. 
> After running for about 43000 timsteps (for each timestep the same 
> computations are done) i get the following error:
> 
> [10.1.1.144:18] Fatal Error: Got an asynchronous event: VAPI_PORT_ERROR 
> (VAPI_EV_SYNDROME_NONE) at line 110 in file viainit.c
> [10.1.1.120:5] Got completion with error, code=VAPI_RETRY_EXC_ERR, vendor 
> code=81, for 10.1.1.144:18
> testNodePoisson2d.Linux.g++.g77.MPI32.ex: viacheck.c:2176: 
> viutil_spinandwaitcq: Assertion `sc->status == VAPI_SUCCESS' failed.
> 
> 
> I would think that this is a bug in my code but given that it does the same 
> thing over and over again for an hour and half before croaking i wouldn't 
> even know where to look for the bug. Anyway, does someone know what this 
> might indicate?
> The MPICH version i am usng is 1.2.5
> 
> thanks
> mat
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss