Thanks Martin! As for Ralph’s question, I was not thinking of deprecating MPI _Comm_spawn and friend for several reasons: We never made headway on a replacement 5-6 years ago in the sessions WG We ought to be able to address the issue with comm
Thanks Martin!
As for Ralph’s question, I was not thinking of deprecating MPI _Comm_spawn and friend for several reasons:
- We never made headway on a replacement 5-6 years ago in the sessions WG
- We ought to be able to address the issue with comm spawn and sessions with relatively small additions to the standard
- This MPI functionality is actually being used widely now and deprecating it would generate major pushback
- DPP could provide functionality to create parent child communicators, although there’s still MPI_Comm_accept/connect as another route if DPP doesn’t
make it into a MPI 6 standard. Note accept/connect isn’t problematic for sessions currently.
We’ll iterate on these
From:
mpiwg-sessions <mpiwg-sessions-bounces@lists.mpi-forum.org> on behalf of MPI Sessions working group <mpiwg-sessions@lists.mpi-forum.org>
Reply-To: MPI Sessions working group <mpiwg-sessions@lists.mpi-forum.org>
Date: Friday, January 24, 2025 at 3:24 AM
To: MPI Sessions working group <mpiwg-sessions@lists.mpi-forum.org>
Cc: Martin SCHREIBER <martin.schreiber@univ-grenoble-alpes.fr>
Subject: [EXTERNAL] Re: [mpiwg-sessions] Topics for Sessions WG meeting 27th January 2025
Dear all, I'll be at a workshop on Monday, hence, it's not sure whether I can attend this meeting. Resonating with Dominik, I'd like to propose that we're working
in the next months on writing down step-by-step the DPP approach in Latex as an
|
This Message Is From an External Sender
|
|
This message came from outside your organization.
|
|
|
Dear all,
I'll be at a workshop on Monday, hence, it's not sure whether I can attend this meeting.
Resonating with Dominik, I'd like to propose that we're working in the next months on writing down step-by-step the DPP approach in Latex as an extension to MPI, how it works, the rationality, etc.
We were working on this over the last years, had many discussions, finally also have an implementation, others also use it, etc. I think it's now time to try to get it into MPI.
In this way we'll structure things nicely, have some concrete discussion basis, etc.
Any feedback is appreciated.
@Howard, if you have some better idea, please let me know.
Cheers,
Martin
On 24.01.25 10:47, Dominik Huber via mpiwg-sessions wrote:
Hi Ralph, in this context, we are discussing a dynamic process approach based on PSets and PSet operations (see Additional PSET query/manipulation functions). The
output of a PSet operation would be a list of PSet IDs that might refer to new
|
This Message Is From an External Sender
|
|
This message came from outside your organization.
|
|
|
Hi Ralph,
in this context, we are discussing a dynamic process approach based on PSets and PSet operations (see Additional PSET query/manipulation functions).
The output of a PSet operation would be a list of PSet IDs that might refer to new processes and can be used to create new MPI groups/communicators.
Regards,
Dominik
Just curious - if you get rid of MPI_Comm_spawn, then how does one generate new processes? I assume some other API must replace it, so mostly wondering what that
API is. FWIW: generating new processes without automatically connecting them to
|
This Message Is From an External Sender
|
|
This message came from outside your organization.
|
|
|
Just curious - if you get rid of MPI_Comm_spawn, then how does one generate new processes? I assume some other API must replace it, so mostly wondering what that API is.
FWIW: generating new processes without automatically connecting them to the parent seems like a reasonable thing to do, so long as you provide back some info (e.g., a group ID or namespace) that allows you to do it later if you so choose.
This Message Is From an External Sender
This message came from outside your organization.
Let me know if you have topics to discuss at our next Sessions WG meeting on Monday 1/27/25.
I’ve attached some slides with ideas for discussion by the WG this calendar year. We can discuss these at the meeting.
|
|
Los Alamos National Laboratory
|
<sessions_wg_01_27_24.pdf>_______________________________________________
mpiwg-sessions mailing list
mpiwg-sessions@lists.mpi-forum.org
https://lists.mpi-forum.org/mailman/listinfo/mpiwg-sessions
_______________________________________________
mpiwg-sessions mailing list
mpiwg-sessions@lists.mpi-forum.org
https://lists.mpi-forum.org/mailman/listinfo/mpiwg-sessions
--
Univ.-Prof. Dr. Martin Schreiber
Applied Mathematics | High Perf. Scientific Comp. for PDEs
Université Grenoble Alpes | Lab. Jean Kuntzmann | Inria | France