[MPICH] wmpiconfig error on windows mpich2-1.0.2: MPICH2 not installed or unable to query the host

[David Ashton]

Aaron,

I suspect the problem is with your firewall.  The way smpd works is that the
client connects to port 8676 and then another smpd.exe is launched in the
context of the client user.  The client then re-connects to this new smpd
process on another port.  Your firewall is probably blocking the second
connection attempt.  The easy solution to this is to not use a TCP firewall.

You will run into problems running an MPI application if you only allow
access to port 8676 and use the default socket channel.  Each MPI process
selects an OS determined port to listen on and these will be blocked by your
firewall.

You can restrict the ports that MPI applications use by specifying the
environment variable MPICH_PORT_RANGE=minport,maxport.  This will work for
smpd also but you will have to get the variable specified before the smpd
service is started by the system.  This probably can be done by setting
MPICH_PORT_RANGE in the system environment variables (make sure the range
includes 8676).

"smpd.exe -status host" is working for you because this mechanism does not
spawn a second smpd process and only needs access to port 8676.

-David Ashton

-----Original Message-----
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Aaron Darling
Sent: Tuesday, July 19, 2005 12:20 PM
To: mpich-discuss at mcs.anl.gov
Cc: Paul Liss
Subject: [MPICH] wmpiconfig error on windows mpich2-1.0.2: MPICH2 not
installed or unable to query the host

Hi All,
I'm attempting to configure mpich2 on a group of windows machines.  We
have automatically deployed the mpich2 msi on several of the machines,
while on two of the machines we manually installed mpich2-1.0.2 so that
we could have the GUI components which deployed installations lack.

After deploying, we updated our firewall policies to enable access to
smpd.exe on port 8676 and running `telnet geb 8676` spits out something
like this to the screen:
C:\Program Files\MPICH2\bin>telnet geb 8676
1.0.2 19169
which I take as being able to successfully connect to the remote smpd.

I'm now trying to use wmpiconfig to configure the daemons.  When I
attempt to read the configuration from the remote host I get the
following message:
GEB: MPICH2 not installed or unable to query the host

However, smpd seems to think it's running on geb:

C:\Program Files\MPICH2\bin>smpd -status geb
smpd running on geb

geb is one of the machines on which mpich2-1.0.2-1 was manually
installed with GUI components.

If I connect to one of the machines that was automatically deployed
using mpich-1.0.1 I see:

C:\Program Files\MPICH2\bin>telnet khufu 8676
5436

but no version number.  Connecting using smpd -status fails:

C:\Program Files\MPICH2\bin>smpd -status khufu
Aborting: unable to connect to khufu

OK, so different versions of mpich won't talk to each other.  I can live
with that.

Why won't wmpiconfig work on my two manually installed mpich2-1.0.2-1 hosts?
Is it a problem that they are in different windows domains even though
they are on the same /24 subnet?

Can anybody give me a clue as to how I can debug the problem or what may
be wrong?

Thanks in advance for your help, hopefully I'm missing something obvious
to you guys!

-Aaron