[mvapich-discuss] Shmem error
Akshay Venkatesh
venkatesh.19 at buckeyemail.osu.edu
Thu Aug 21 17:36:54 EDT 2014
Katherine,
The following problem:
*shmem open failed for
file:/dev/shm/slot_shmem-coll-kvs_236134_0-udc-ba38-4d-0-1614.tmp*
most likely indicates that there was no write permission on /dev/shm or
that /dev/shm has limited space.
The subsequent problem:
*collective shmem allocation failed: No such file or directory*
occurred because
*file:/dev/shm/slot_shmem-coll-kvs_236134_0-udc-ba38-4d-0-1614.tmp
*was never created
Could you check with permissions and space availability first and then
retry? Additionally, it is recommended that the newer MVPAICH2-2.0 be
used. Download link is available at (http://mvapich.cse.ohio-state.edu)
On Thu, Aug 21, 2014 at 5:06 PM, Katherine Holcomb <
kah3f at eservices.virginia.edu> wrote:
> In trying to prepare a new system with OFED we have a test code that fails
> under MVAPICH2 1.9 on ALLREDUCE with the following error:
>
> [udc-ba38-4d:mpi_rank_0][mv2_shm_coll_init] shmem open failed for
> file:/dev/shm/
> slot_shmem-coll-kvs_236134_0-udc-ba38-4d-0-1614.tmp
>
> [cli_2]: [cli_0]: aborting job:
> Fatal error in PMPI_Reduce:
> Other MPI error, error stack:
> create_2level_comm(885): collective shmem allocation failed: No such file
> or directory
>
> (one for each rank).
>
> The same code with the same inputs works fine under OpenMPI. It also
> works at a different site with MVAPICH2 1.9a2.
>
> I am not even sure where to start to debug this.
>
> --
> Katherine Holcomb
> UVACSE kholcomb at virginia.edu
> 112 Albert Small Building (434) 982-5948
> University of Virginia Charlottesville, VA 22904
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
--
- Akshay
http://www.cse.ohio-state.edu/~akshay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20140821/ff3af1e7/attachment.html>
More information about the mvapich-discuss
mailing list