[MPICH] Tools to play(debug/monitor) with source - a developers perspective ?

Rajeev Thakur thakur at mcs.anl.gov
Tue Jun 14 08:57:01 CDT 2005


You can also use the script mpigdb, which serves as a parallel interface to
gdb. Info on how to use it is in the Users guide and in src/pm/mpd/README.

Rajeev 

> -----Original Message-----
> From: owner-mpich-discuss at mcs.anl.gov 
> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Anthony Chan
> Sent: Monday, June 13, 2005 11:57 PM
> To: Syed Irfan
> Cc: mpich-discuss
> Subject: Re: [MPICH] Tools to play(debug/monitor) with source 
> - a developers perspective ?
> 
> 
> 
> On Tue, 14 Jun 2005, Syed Irfan wrote:
> 
> > What tools are used to modify/debug/analyze/monitor mpi 
> source (API) ?
> 
> I am not sure what you mean by mpi source!  Here is what I know.
> 
> For debugging, you could use gdb/ddd (good for few processes)
> 
> mpiexec -n 2 ddd a.out
> 
> This is only possible with mpich2 and not mpich1.  For complicated
> or large job, you may need totalview (I haven't tried this with mpich2
> yet, but the support is there).
> 
> 
> For performance tunning:
> 
> --At user level, you could MPE's logging capability to profile
>   your mpi code, i.e. you first generate a clog2 file by linking
>   your code with mpe libraries and convert the file to slog2 file
>   and view it in jumpshot.  I believe you can use TAU from 
> university of
>   oregon.
> 
> --At developer level, you could configure MPICH2 with rlog support.
>   Running mpi code will generate a rlog file which contains 
> the details
>   of mpich2's internal.  Again you can view rlog file with jumpshot.
> 
> > An insight into the procedure or a blog from developers 
> perspective on
> > his activity to improve/modify mpich source would be great :)
> 
> May not be what you are looking for, but our cvs repository has been
> posted on our website:
> 
> http://www-unix.mcs.anl.gov/~gropp/projects/parallel/MPICH/mpi
> ch2/changes/index.html
> and
> http://www-unix.mcs.anl.gov/mpi/mpich2/cvsweb.cgi/
> 
> A.Chan
> 
> 




More information about the mpich-discuss mailing list