I'm afraid that we have done no verification of LAM under cygwin; I
know so little about Windows that I wouldn't even know where to
start. Sorry! :-(
Perhaps some others here on the list might have some ideas...?
On Oct 10, 2008, at 12:54 PM, Gustavo Seabra wrote:
> Hi All,
>
> I just installed LAM/MPI (Lam 7.1.3) in my computer, using Cygwin and
> g95. The installation seems to have completed normally, although for
> some reason LAM keeps creating a 'libtool' executable in the
> installation root directory, even though libtool is installed in my
> system: ("$ which libtool" returns "/usr/bin/libtool").
>
> However, when I try to use LAM, I get this "bad file descriptor"
> problem whenever I execute a *second* mpirun. (see the example below).
> This happens when using the same program twice or using a second
> (different) program. The first execution is always OK, the second
> crashes the LAM environment.
>
> If, instead, I start LAM with 'lamboot -d', some programs (e.g.
> trivialf.exe) can execute multiple times with no problems. But others
> still crash in the second execution. (The program I originally want to
> use LAM with - AMBER - crashes on the second execution, just like
> before). The 'hello.exe' example also crashes after the second
> attempt.
>
> Just some extra information that may (or may not) be relevant:
> 1. I installed LAM under a regular user account. However, all the
> executables are in a directory in my account, and I have full rights
> to all files.
>
> 2. During the installation, the Windows Firewall blocked "conftest",
> but installation did not stop and went all the way to the end, no
> problems.
>
> 3. The first time I used LAM, again the windows firewall blocked
> "lamd" and "lamboot". After that, I manually added exceptions to lamd
> lamboot, so it doesn't complain about it anymore. The firewall keeps
> blocking any program that ses LAM, but the execution still runs to the
> end with no problems (at least the first time, as I reported)
>
> This blocking from the Windows firewall is unlikely to be the
> problem: I created an exception for "sander.MPI" (the executable for
> the "AMBER" program) so that, when I use amber I now get no messages
> from the firewall. Still, the same problem happens (it can be executed
> only once).
>
> 4. I also tried issuing 'lamclean' between executions, doesn't help
> (nothing changes).
>
> If there is any more information that can be helpful in tracking this
> down, please let me know.
>
> What can be causing this? Any suggestions?
>
> Thanks a lot.
> Gustavo.
>
> Example: This is representative of the error I'm seeing.
> =========================================
> $ lamboot
>
> LAM 7.1.3 - Indiana University
>
> $ mpirun -np 2 trivialf.exe
> rank 1 received message
> rank 0 sent message
> $ mpirun -np 2 trivialf.exe
> rank 1 received message
> rank 0 sent message
> lamd kernel: problem with select() (1): Bad file descriptor
> =========================================
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
--
Jeff Squyres
Cisco Systems
|
