[mpich-discuss] Windows/Linux MPICH2

[jayesh at mcs.anl.gov]

Hi,
 
 Couple of suggestions,

1) Turn SMPD dynamic host option off on the unix machines (Set the env variable "setenv SMPD_OPTION_NO_DYNAMIC_HOSTS 1" before starting the process manager, "smpd -s", on the unix machines.)

2) Do you have the same version of cpi installed on both unix and windows machines (The unix machines typically compile the non-interactive version, cpi.c and the windows installation have the interactive version, icpi.c) ?

Regards,
Jayesh

----- Original Message -----
From: "Matthew Chambers" <matthew.chambers at vanderbilt.edu>
To: mpich-discuss at mcs.anl.gov
Sent: Thursday, March 18, 2010 12:55:22 PM GMT -06:00 US/Canada Central
Subject: Re: [mpich-discuss] Windows/Linux MPICH2


Hi Jayesh, 

I reconfigured with ch3:sock and I get the same hanging problem. Testing with cpi: 

../bin/mpiexec -hosts 2 <centos-hostname> <windows-hostname> -path "/hactar/src/mpich2-1.3a1/examples;C:\program files (x86)\mpich2\examples" cpi 
User credentials needed to launch processes on deepthought: 
account (domain\user): root (yes, the windows username is root) 
password: 
Process 0 of 2 is on jeltz.mc.vanderbilt.edu 
pi is approximately 3.1415926544231318, Error is 0.0000000008333387 
wall clock time = 0.000766 
<hangs> 

Windows firewall is disabled and I get the same results whether the CentOS box's iptables is on or off. 

Thanks, 
-Matt 

On 3/18/2010 12:16 PM, Jayesh Krishna wrote: 

Hi,
 Yes, MPICH2 currently does not support heterogeneous machine configurations (You cannot run your MPI program across 32-bit/64-bit machines/data-models). You should use IA32 version of MPICH2 on the windows 7 machine.
 Also make sure that you have configured the unix boxes to use sock (--with-device=ch3:sock) as the channel (the default is nemesis).
 Make sure that you don't have firewalls running on any machines.
 Let us know the results.

Regards,
Jayesh
----- Original Message -----
From: "Matthew Chambers" <matthew.chambers at vanderbilt.edu> To: mpich-discuss at mcs.anl.gov Sent: Thursday, March 18, 2010 11:30:48 AM GMT -06:00 US/Canada Central
Subject: [mpich-discuss] Windows/Linux MPICH2


Hi, 

I'm trying to get a cluster working with three types of nodes: 64-bit CentOS 5.4, 64-bit Windows 7, and 32-bit Fedora 8. The 32-bit machines can't run a 64-bit OS so I'm stuck with 32-bit. I'm trying to do this with SMPD but having trouble. Am I correct in thinking that I have to use/compile the 32-bit version of MPICH2 on the 64-bit machines? 

It appears that I can't use the Windows node as the typical head node because it needs to map a network drive with different user credentials in order to run. So I'm back to using the CentOS box as a head node. However, when I try to run a job which includes the Windows node, the job hangs. 

"smpd -status <windows-hostname>" works as expected from the CentoOS box 
"smpd -status <centos-hostname>" works as expected from the Windows box 

But trying to run a job and it hangs after outputting "Process #1 (<windows-hostname>) has started." 

I've attached the SMPD debug logs for both the CentOS and Windows boxes. I've tried both 1.2.1p1 and 1.3a1 with the same result. 

Thanks, 
Matt 

_______________________________________________
mpich-discuss mailing list mpich-discuss at mcs.anl.gov https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss 
_______________________________________________
mpich-discuss mailing list
mpich-discuss at mcs.anl.gov
https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss