(this thread started as a LAM question [http://www.lam-mpi.org/MailArchives/lam/2006/06/12497.php], and one message contained an Open MPI question, so I took the liberty of moving it to the OMPI user's list)
> As for openmpi, I get a lot of messages like this > > global_ssi(1441) malloc: *** Deallocation of a pointer not malloced: > 0xbffe9e70; This could be a double free(), or free() called with the > middle of an allocated block; Try setting environment variable > MallocHelp to see tools to help debug > global_ssi(1441) malloc: *** Deallocation of a pointer not malloced: > 0x1; This could be a double free(), or free() called with the > middle of > an allocated block; Try setting environment variable > MallocHelp to see > tools to help debug > > and then it crashes randomly with a seg fault (no error message). Yikes. Note that these are messages from OSX itself, not Open MPI. It doesn't necessarily mean anything (i.e., the error could still be in your code or in Open MPI), but it is certainly an indication of Badness. What version of Open MPI did you try? We just released v1.1 last week -- could you give that a whirl if you haven't already? I'm also not an OSX expert, but I do know that setting some environment variables will cause those messages to be more verbose (potentially dumping a stack trace or something...? I honestly don't remember offhand). If you have the cycles could you set that MallocHelp environment variable and try to track down where the error is occurring? It would be good to know, for example, if it's happening in your application or in Open MPI. From there, we can try to back-track and see where the original problem is located. Thanks. -- Jeff Squyres Server Virtualization Business Unit Cisco Systems
