Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

From: Reuti (reuti_at_[hidden])
Date: 2007-09-17 16:03:18


Am 17.09.2007 um 21:33 schrieb Josh Hursey:

> Sorry for catching up to this thread a bit late.
> In the Open MPI trunk there is a mpirun option to preload a binary
> before execution on remote nodes called '--preload-binary' (or '-s').
> It has been tested with many of the resource managers (and should be
> fairly resource manager agnostic), but I don't believe it has been
> tested in the Xgrid environment. It will not use the Xgrid
> distribution mechanism, but rsh/ssh transfer mechanisms. Native Xgrid
> support can most likely be supported, but I do not know of any
> developer planning to do so at the moment.
> It has been a while since I tested the '--preload-binary' option to
> mpirun so please let us know if you experience any problems. There is
> also a '--preload-files=' option to mpirun that will allow arbitrary
> files to be distributed to remote machines before a job starts as
> well. Both options are useful when working in non-shared file systems.

this is fantasic - but is this a hidden feature, compile-time option,
or lack of documentation? When I issue "mpirun -help" I don't get
these options listed. Hence I wasn't aware of them.

-- Reuti

> Cheers,
> Josh
> On Sep 17, 2007, at 3:17 PM, Reuti wrote:
>> Am 17.09.2007 um 16:34 schrieb Brian Barrett:
>>> On Sep 10, 2007, at 1:35 PM, Lev Givon wrote:
>>>> When launching an MPI program with mpirun on an xgrid cluster, is
>>>> there a way to cause the program being run to be temporarily
>>>> copied to
>>>> the compute nodes in the cluster when executed (i.e., similar to
>>>> what the
>>>> xgrid command line tool does)? Or is it necessary to make the
>>>> program
>>>> being run available on every compute node (e.g., using NFS data
>>>> partions)?
>>> This is functionality we never added to our XGrid support. It
>>> certainly could be added, but we have an extremely limited supply of
>>> developer cycles for the XGrid support at the moment.
>> I think this should be implemented for all platforms, if it would
>> have to be part of OpenMPI at all (the parallel library Linda is
>> offering such a feature). Otherwise the option would be to submit the
>> job using XGrid, or any another queuingsystem, where you can setup
>> such file tranfers in any prolog script (and epilog to remove the
>> programs again) - or copy it to the created $TMPDIR which I would
>> suggest if you decide to use e.g. SUN GridEngine, as this will be
>> ereased after the job automatically.
>> But just for curiosity: how is XGrid handling it, as you refer to
>> command-line-tool? If you have a jobscript with three mpirun-commands
>> for three different programs, XGrid will transfer all three programs
>> to the nodes for this job, or is it limited to be just one mpirun is
>> just one XGrid job?
>> -- Reuti
>>> Brian
>>> --
>>> Brian W. Barrett
>>> Networking Team, CCS-1
>>> Los Alamos National Laboratory
>>> _______________________________________________
>>> users mailing list
>>> users_at_[hidden]
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
> _______________________________________________
> users mailing list
> users_at_[hidden]