Hi all, as Martin mentioned correctness tools like Marmot will benefit from the extra interfaces. Especially the query interface for MPI handles sounds very promising, as it would simplify the surveillance of correct handle usage. So the Marmot world is listening, unfortunately I oftentimes have a conflict at the telco time, so I am not always at the telcos. -Tobias Martin Schulz wrote:
On Sep 2, 2009, at 1:46 PM, Marc-Andre Hermanns wrote:
Hi Martin,
The goal of this working group is define interfaces that can be used by both performance and correctness tools to gain information about internals of the MPI library, interactions between an application and the MPI library, as well as the system environment an MPI application runs in. This is intended as an extension to the existing and widely used PMPI interface. These efforts will help in providing reliable and portable interfaces for MPI tools with new functionality currently not covered by the MPI standard.
Does 'correctness' tools include debuggers? It immediately triggered things like "Marmot" and not directly Totalview, DDT, etc., as I would usually not refer to a debugger as a correctness checker (but then, I am not a native speaker).
This was intended to cover both types of tools: correctness checkers and debuggers.
To make that a little clearer I would propose the following change:
The goal of this working group is to define interfaces that can be used by third party tools to obtain internal information about the MPI library, as well as the system environment an MPI application runs in. These tools include performance analysis tools, correctness checkers and debuggers. These interfaces are complementary to the existing PMPI interface. These efforts will help in providing reliable and portable interfaces for MPI tools with new functionality currently not covered by the MPI standard.
Works for me.
In particular, this WG is aiming at the standardization of the following interfaces: - A scalable version of a process acquisition interface to identify and locate all processes which are part of an MPI job. - An interface to look for and locate tool DLLs. - An interface to inspect message queues. - An interface to understand progress in collective communication. - An interface to query additional semantic information for opaque MPI handles. - A performance information interface that allows MPI implementations to export additional low level performance information including, but not limited to, the operating state of the MPI library.
The time frame that was discussed today for the MPI 3.x effort was 3+1 years. That is 3 years until 3.0 is ready, and an additional year for something like 3.1 (Jeff, feel free to correct me on this). I don't have the experience yet, to estimate whether it is doable in this time frame, but the list seems to be quite long. All the items on the list seems valid points of interest, though.
Most of these items have already been discussed and we have concrete proposals on the table and they are in the process of being written up. The only remaining item is the process acquisition interface. Nevertheless, this is a lot of work and I am worried about this new March deadline for getting this done to be included in the MPI 3 standard. We should be able to get some of these items through, but it will a lot of work to make this work for all items (or even just the ones for which we have concrete ideas), in particular, since this schedule seems not to allow for any substantial feedback from the forum - basically, it seems that if there are even small objections that require a rewrite, the proposal already ends up in 3.1 or later. Has this been discussed?
Martin
Best regards, Marc-Andre -- Marc-Andre Hermanns Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH D-52425 Juelich Germany
Phone : +49-2461-61-2054 Fax : +49-2461-61-6656 eMail : [email protected] WWW : http://www.fz-juelich.de/jsc/
JSC is the coordinator of the John von Neumann Institute for Computing and member of the Gauss Centre for Supercomputing
Sitz der Gesellschaft: Juelich Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 Vorsitzende des Aufsichtsrats: MinDir'in Baerbel Brumme-Bothe Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender), Dr. Ulrich Krafft (stellv. Vorsitzender), Prof. Dr. Harald Bolt, Prof. Dr. Sebastian M. Schmidt _______________________________________________ Mpi3-tools mailing list [email protected] http://*lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-tools
_______________________________________________________________________ Martin Schulz, [email protected], http://people.llnl.gov/schulzm CASC @ Lawrence Livermore National Laboratory, Livermore, USA
_______________________________________________ Mpi3-tools mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-tools
-- Tobias Hilbrich GWT-TUD GmbH