[mpich-discuss] Strange MPICH2 behavior in Clearcase

[chong tan]

There a quite some issues with running MPICH2 in Clearcase.  At first, I thought I can run the daemon from outside
a view as I always run mpiexec from a view.  That turned out to be a problem as the executables inside a view somehow
became invisible.  


I don't use mpdboot because of ssh.  I have a few schemes that I use depending on a grid's setting.  A basic scheme
is to limit each node on the grid to run 1 job, which is a script that launch mpd at the beginning, and kill it at the end.

In a simple word,  I can't use anything that requires login on grid nodes.

tan



________________________________
From: Pavan Balaji <balaji at mcs.anl.gov>
To: mpich-discuss at mcs.anl.gov
Sent: Mon, April 12, 2010 2:41:59 PM
Subject: Re: [mpich-discuss] Strange MPICH2 behavior in Clearcase


How are you launching the mpd daemons? mpdboot would need to ssh into the machines to launch mpd daemons as well. If ssh works fine for you, Hydra should work correctly as well. What's the problem you are seeing?

-- Pavan

On 04/08/2010 03:22 PM, chong tan wrote:
> hydra is not a solution I can use.  The main issue is that login is not allow in most grid env.
> 
> running mpd outside of clearcase view does not work as the executables are inside views.  It looks like
> the problem could be inside mpd or mpiexec.  The likely cause is that I am running from 2 different views.
> It would be nice if I can run 2 mpd, one per view, on the same machine, to help isolate the problem.
> 
> tan
> 
> 
> ------------------------------------------------------------------------
> *From:* Dave Goodell <goodell at mcs.anl.gov>
> *To:* mpich-discuss at mcs.anl.gov
> *Sent:* Thu, April 8, 2010 10:29:40 AM
> *Subject:* Re: [mpich-discuss] Strange MPICH2 behavior in Clearcase
> 
> We (the MPICH2 developers) don't have any experience with clearcase.  Perhaps someone in the community does though.  Because of the way that clearcase works, strange filesystem behavior in MPD doesn't surprise me.
> 
> I would try two things:
> 
> 1) Use hydra with the "-wdir" option instead of MPD, as we discussed in a previous thread: https://lists.mcs.anl.gov/mailman/htdig/mpich-discuss/2010-January/006356.html
> 
> 2) Run your mpdboot from a directory outside of your clearcase mount.  This might avoid some of these problems.
> 
> -Dave
> 
> On Apr 8, 2010, at 11:59 AM, chong tan wrote:
> 
>  > I am running MPICH2 1.2.1 in Clearcase and observed this crazy MPICH2 behavior.  basically,
>  > I have 2 code streams (or clearcase view), one is a stable one and is where I am writing new code.
>  > the 2 streams are not compatible to some extend.  Say my executable is called MY, so there are 2
>  > version of it.
>  >
>  > I launch MY in a common script using :
>  >
>  > mpiexec MY < options> : MY < options>  : ...
>  >
>  > I run both, alternatedly on my test cases.  the run dir is cleaned to the bone between the runs.  From time
>  > to time, I run into situation where the wrong executable was picked up.  the problem goes away after
>  > I restart mpd.
>  >
>  > Question:  is this a know issue of MPICH2 ? or this is a issue of MPICH2 + clearcase ?
>  >
>  > thanks
>  > tan
>  >
>  >
>  >
>  >
>  > _______________________________________________
>  > mpich-discuss mailing list
>  > mpich-discuss at mcs.anl.gov <mailto: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 <mailto: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

-- Pavan Balaji
http://www.mcs.anl.gov/~balaji
_______________________________________________
mpich-discuss mailing list
mpich-discuss at mcs.anl.gov
https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss



      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20100412/cb39da7a/attachment.htm>