[mpich-discuss] MPICH2 mpiexec console window on XP?

[Jayesh Krishna]

Hi,
 It was my mistake. You have to set the PMI_RANK env variable to 0 (not 1).

Regards,

Jayesh


  _____  

From: Raymond L Swartz [mailto:Raymond.L.Swartz at aero.org] 
Sent: Tuesday, April 15, 2008 9:25 AM
To: Jayesh Krishna
Subject: RE: [mpich-discuss] MPICH2 mpiexec console window on XP?



Thank you for your quick reply. 

I just tried as you suggested, and when I issued the run command nothing
happened.  I didn't get the console window but I also did not have the
program start up. 

By looking at the proces window I see the program starts up very briefly and
then shuts down.  No error is generated.  I'm currently looking for what is
happening with the debugger. 

Without these settings the program starts up with the new console window. 

Is there a mistake somewhere in the instructions below? 

Is this the direction that the MPI project is going in the future, where the
user has to do special steps to get basic single-processor operation?  Our
application had (with the previous library) been something that the basic
'clueless user' could use in single-processor mode without ever knowing it
had MPI capability, but for the saavy they could fire it up with mpiexec and
be off to the races with parallel processing.  The 'clueless user' I'm
talking about doesn't even know what an envirfonment variable is, not to
mention how to set it.  I'm concerned about how we keep operating in that
model with the new library. 

Thanks, 
Ray 




"Jayesh Krishna" <jayesh at mcs.anl.gov> 


04/14/2008 02:59 PM 


To
Raymond L Swartz <Raymond.L.Swartz at aero.org> 

cc
<mpich-discuss at mcs.anl.gov> 

Subject
RE: [mpich-discuss] MPICH2 mpiexec console window on XP?

	




Hi, 
 The new behaviour that you see is due to the support for "Singleton Init"
processes that was added in 1.0.7 (It is an expected behaviour). Currently
there is no way to "turn off" the feature. 
 You can start an MPI process without the use of a process manager by
specifying the PMI environment variables by hand (After the PMI env
variables are set the MPI library uses the values provided and does not try
to connect to the existing process manager, smpd, using the job launcher,
mpiexec.). 
  
########### PMI vars to be set for single process MPI jobs with no PM
########### 

set PMI_ROOT_HOST=%COMPUTERNAME% 


set PMI_ROOT_PORT=9222 


set PMI_ROOT_LOCAL=1 


set PMI_RANK=1 


set PMI_SIZE=1 


set PMI_KVS=mpich2 


####################################################################### 
  Please note that you should unset the variables before running other MPI
jobs. 
  We will fix our code in the next release so that users don't see the
console window of mpiexec when running their applications (GUI apps). For
now, please follow the instructions above to avoid automatic launching of
the job launcher. 
  
(PS: This support allows a singleton init MPI process (MPI process started
without a job launcher)  to function as a normal MPI process. eg: Without
this support a singleton init MPI process cannot spawn other MPI processes
(MPI_Comm_spawn()). 
  
Regards, 
Jayesh 


  _____  

From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Raymond L Swartz
Sent: Monday, April 14, 2008 4:25 PM
To: mpich-discuss at mcs.anl.gov
Subject: [mpich-discuss] MPICH2 mpiexec console window on XP?


Hi, 

I've just updated to MPICH2 1.0.7 for a project we have here (which has been
running with the previous version just fine) and notice some new behavior. 

When the job starts up now, in addition to the standard GUI we expect, we
are seeing a 'console ' window with the title C:/Program
Files/MPICH2/bin/mpiexec.exe. 

This window does not show up if we launch the executable as 'mpiexec
-localonly 2 <executable>'.  It only appears if we launch the executable
without the mpiexec invocation '<executable>' (nominally, to run as a
single-process job). 

This is new behavior.  Previously we never saw this window.  The console
window is blank, and nothing appears in it. 

Is there a compile-time way to turn it off? 

If it matters, we are building the application in VC++, and using Qt4 as the
GUI library. 

Thanks, 
Ray Swartz 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20080415/0711e089/attachment.htm>