On Feb 18, 2005, at 10:20 AM, Yu Chen wrote:
> Hi, I have been struggling about weeks. Still no clue. Hope somebody
> can help me out :-(
>
> I am using LAM-7.1.2.b13, compiled with Intel-8.1 compiler(both
> CC=icc, CXX=icpc, FC=ifort) withouth problem and installed fine.
>
> We have a 20 nodes cluster, all have 2 AMD Athlon MP1900+ cpus,
> running RHEL-AS-3. Same LAM are installed on each of nodes. Data and
> our program are nfs auto-mounted on each nodes.
>
> After I start the program, it reads some data in, then stoped right
> before the real calculation begins. No errors/warnings, just sit.
>
> Strange thing is then I run "mpitask", it gives out this:
MPITask is showing that you have 8 processes that LAM thinks are still
running. Seems to be exactly what you describe. Because your
application is fortran, it is not possible to show the same output as
top (we don't have argv[0]).
> There are only one process actually running (cyanaexe.intel-), and
> %CPU should be around 50% for the program, right? since only one cpu
> being used as showen at the top, seems they are not communicating.
Again, looks right to me. cyanaexe.intel- is using 100% of one CPU.
top on Linux has kind of a weird notation for per-process CPU time.
100% means using 100% of 1 cpu. So a multithreaded application could
achieve 200% CPU time on a dual processor machine. Not exactly
intuitive, but that's the way they do it.
As to your seeming "hang", I would recommend using gdb to attach to the
running process and see where it is spending its time. Without more
information, I can't offer much more advice than that. You might want
to take a look at the debugging section of the LAM FAQ:
http://www.lam-mpi.org/faq/
Hope this helps,
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have an LAM/MPI day: http://www.lam-mpi.org/
|
