[Mpi3-tools] Next TelCon 1/12 8am PT, 11am ET, 5pm MEZ

[Martin Schulz]

Hi John,

>* I think the current MPIR description on the Wiki needs a lot more 
>detail. I think it needs to include a "C" language description that 
>shows the declarations of the structure types, variables, and 
>functions. It's missing type and variable declarations. At the next

Agreed. We need to add that. Do you agree with the general workflow,
though? Is this how it works in TV?

>  higher level of detail, it needs to elaborate on the semantics of 
> each of the variables and functions, both from the debugger's and 
> MPI perspective; it currently very thin. At the highest level of 
> detail, it should talk about the different "modes" of useing the 
> interface. For example, the "rank 0 process launches" model used by 
> MPICH 1, vs. the "separate mpirun starter process launches" model 
> used by most other MPI implementations.

Shouldn't the interface hide this from the tool? The only piece of
information that should be needed is the initial process the tool
needs to attach to. Enumerating the modes could result in leaving
out implementations that we are not aware of.

>  Also, there are special conventions and  used on various MPI 
> implementations. For example, Blue Gene (and now Open MPI) support 
> the notion of co-spawning tool daemons, and several implementations 
> support the notion of partial (scalable subset) attach.

Yes, we need to document those. We talked about this during the forum,
but nobody there had enough information at hand to list all of these
extensions. We know about the BG/L ones, but we weren't sure about what
else is there. If you have this information, it would be great if you
could add this to the Wiki.


>* On Monday, I think we need to figure out who's going to be the 
>principle author of the document and who is going to edit/maintain 
>the document.

If there is no other volunteer (is there :)?), I can be the central
person to collect the information and to write and maintain the
document. However, the information must come from the MPI implementors
as well as the tools that already use the interface. I also contacted
the O|SS developers, who are using the interface, but most of the
usage is probably in TV and hence it would be great if someone from
TVT could document all the special operations you had to implement
for different MPI implementations.

Martin



>Cheers, John D.
>
>
>Martin Schulz wrote:
> > Hi all,
> >
> > Just as a reminder: we have our next MPI Tools WG TelCon on Monday
> > 1/12 at the usual time at 8am PT, 11am ET, 5pm MEZ. Here is the
> > dial-in information:
> >
> >> Toll Free Dial In Number for Offsite Participants:  866-914-3976
> >> International Access/Caller Paid Dial In Number:  925-424-8105
> >> Dial In Number for LLNL Onsite Participants:  4-8105
> >>
> >> Access Security Code: 860609 #
> >
> > Unfortunately, I won't be able to be there for the whole TelCon
> > since I will try to call-in from on the road. Bronis has agreed
> > to lead the TelCon instead. The main topics will be to continue
> > the discussion of the current state of the MPIR interface as well
> > as a presentation by Chris on the MPIR-2 interface, which includes
> > dynamic process support.
> >
> > Please also take a look at the current MPIR description on
> > the Wiki and suggest changes or corrections. The plan is to
> > convert the Wiki in the coming weeks into a standalone document
> > that we can reference in the MPI 2.2 standard. In particular,
> > we still need to document existing, platform specific extensions
> > to MPIR. If you are aware of such an extension, please add it to
> > the Wiki page.
> >
> > Thanks,
> >
> > Martin
> >
> >
> >
> > _______________________________________________________________________
> > Martin Schulz, schulzm at llnl.gov, http:// people.llnl.gov/schulz6
> > CASC @ Lawrence Livermore National Laboratory, Livermore, USA
> > _______________________________________________
> > Mpi3-tools mailing list
> > Mpi3-tools at lists.mpi-forum.org
> > http:// lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-tools
> >
> >
>_______________________________________________
>Mpi3-tools mailing list
>Mpi3-tools at lists.mpi-forum.org
>http:// lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-tools

_______________________________________________________________________
Martin Schulz, schulzm at llnl.gov, http://people.llnl.gov/schulz6
CASC @ Lawrence Livermore National Laboratory, Livermore, USA