[mpich-discuss] Getting started in Win XP

[Charles Lawson]

I downloaded MPICH2 for Windows in Nov, 2009. I installed in on two PC's
that I call Issac and Gauss that have access to my home wireless network.
The installation file is mpich2-1.2-win-ia32.msi. Issac uses XP-Pro and
Gauss uses XP-Home.

I purchased the Intel Fortran compiler, Vers 11.1, in July, 2009, for use
on Issac, and I regularly use it there. I compiled the test case, fpi.f, on
Issac. I (eventually) got a successful execution on a single PC by using the
statement:

mpiexec -hosts 1 Issac 1 fpi.exe

I issued this command in a directory that contained mpiexec.exe and fpi.exe.
I am the only user of this PC, and I do not use a password -- maybe I should
say I have a blank password. Anyway the above command line didn't work
initially because it asked for a password, and I didn't have one. I then set
up a second User Name and gave it a password. Working under that User Name I
again issued the above command line and I gave it the password when it asked
for it, and it worked!!  I was happy.

Moving on to try running this on my two PC's, I tried the command line:

mpiexec -hosts 2 Issac 1 Gauss 1 fpi.exe

I had Gauss running with a User Name that has a password (which may or may
not have any bearing on the situation.)

This failed with the message:

launch failed: CreateProcess(fpi.exe) on 'Gauss' failed, error 2 - The
system cannot find the file specified.

Then I tried giving fpi.exe a full path. I used:

mpiexec -hosts 2 Issac 1 Gauss 1 \\Issac\mpi\fpi.exe<file://issac/mpi/fpi.exe>

That got  further, but still not a complete execution. This responded with:
   Process 0 of 2 is alive
   Enter the number of intervals: (0 quits)
   Process 1 of 2 is alive

I entered 50, but there was no response. I entered 0 and still no response,
so I killed the run.

I added four Fortran write statements into the test case, fpi.f. Before and
after the Call MPI_BCAST and before and after the Call MPI_REDUCE. Each
write included the value of myid in the output. I recompiled that. I then
tried the last command line above again. It responded with:

   Process 0 of 2 is alive
   Enter the number of intervals: (0 quits)
   Process 1 of 2 is alive
   Proc 1 before call MPI_BCAST  <= My debug write.
   50                                           <= My entry from the
keyboard.
   Proc 0 before call MPI_BCAST  <= My debug write.

and then just hung at that stage.

I conclude that fpi.exe is executing on both PC's, and on both PC's it has
successfully gotten through calls to MPI_INIT, MPI_COMM_RANK and
MPI_COMM_SIZE.  The "print ... alive" statement functioned correctly on both
PC's. Execution continued correctly up to the Call to MPI_BCAST. There it
died.  Why?

I will greatly apprciate suggestions as to how to get this test case to run
correctly. I am well acquainted with Fortran programming, but not well
acquainted with lower level operating system or networking components.

Hoping for help,

Chuck
=====
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100421/151b8717/attachment.htm>