Re: kernel related MPI problem?
On Friday, Nov 21, 2003, at 20:56 Australia/Sydney, Peter Klaver wrote: Hi everyone, A collegue in my lab, knowledgeable in unix but not specifically in Darwin, wants to run MPI code on a dual G5, but there some problems when using the smp-compiled libraries (-comm=share). He suspects it's a kernel-related matter, see below. From his description, can anyone see if this is indeed the case? If anyone on this list happens to have smp-MPI running ok, then any other relevant info is of course also much welcomed. kind greets, Peter Shared memory problem. During testing the smp-MPICH software the following error occurs: P4xxxxx shmget error -1 This looks like a shmget error. Testing the operating system with: sysctl -a |grep sysv kern.sysv.shmmax: 4194304 kern.sysv.shmmin: 1 kern.sysv.shmmni: 32 kern.sysv.shmseg: 8 kern.sysv.shmall: 1024 Looks allright to me. I have tested the shmget call by the following C-program. #include <sys/types.h> #include <sys/ipc.h> #include <sys/shm.h> int main() { key_t key; int size; int shmflg; int shmid; size = 100; key = IPC_PRIVATE; shmflg = IPC_CREAT & IPC_EXCL & (7+7*8+7*64); shmflg = shmflg&IPC_CREAT; Don't you really mean: shmflg = IPC_CREAT | IPC_EXCL | 0777; Do you? shmid=shmget(key, size, shmflg); printf("result %d/ \n",shmid); return 1; } It always returns -1. Runnig the C-program on other (non G5) platforms returns a normal shmid. my hardware platform is dual-G5 my OS version is 10.3 or 10.2.8, similar problems on both. Worked fine for me on 10.2.7 and 10.3. Remember: - those shm segments are persistent. They'll have to be manually deleted. - there's only 4 MB available. - you might have run out of space! - you're really better off using mmap(2) under Darwin. Cheers, -- Paul Ripke Unix/OpenVMS/TSM/DBA I love deadlines. I like the whooshing sound they make as they fly by. -- Douglas Adams _______________________________________________ darwin-kernel mailing list | darwin-kernel@lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.
participants (1)
-
Paul Ripke