I am testing the 2.0.0rc3 tarball PLUS the patch from PR1232 to fix the dependence on 64-bit atomics. On an ARM system with only 256MB of memory, I am seeing the following failure which did NOT occur in my testing of 1.10.3rc4.
$ mpirun -mca btl sm,self -np 2 examples/ring_c' [armel-jessie:03953] create_and_attach: unable to create shared memory BTL coordinating structure :: size 134217728 -------------------------------------------------------------------------- It appears as if there is not enough space for /run/user/19214/openmpi-sessions-19214@armel-jessie_0/34799/1/shared_mem_pool.armel-jessie (the shared-memory backing file). It is likely that your MPI job will now either abort or experience performance degradation. Local host: armel-jessie Space Requested: 134217736 B Space Available: 25968640 B -------------------------------------------------------------------------- [...] Your MPI job is now going to abort; sorry. [...] I already determined that changing "-mca btl sm,self" to "-mca btl tcp,self" works, but would rather not go that route if "sm" can be preserved. So, if there are other MCA parameters I could/should be setting to avoid this please let me know. If this system is now "officially too small" to run btl:sm, then I will just have to accept that. However, I wanted to bring this to the attention of the devel list in case this change in behavior relative to v1.10 was *not* expected behavior. BTW: The awkward line break between "backing" and "file" is in the original output (not an artifact of my email). -Paul [who still remembers owning an i486 with a 40MB hard drive] -- Paul H. Hargrove phhargr...@lbl.gov Computer Languages & Systems Software (CLaSS) Group Computer Science Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900
