- mpiwg-ft - mailman3.cels.anl.gov

Meeting Reminder, Thursday July 16th @2 ET
by Whitlock, Matthew Joseph 15 Jul '26

15 Jul '26

Current agenda: * Form working group consensus on reasons for and against extending the fault model beyond hard process failure * Agree on a name for a simplified `MPI_Comm_agree(comm, &agreed_failed_group)` function and decide on who wants to write the first draft * Begin edits to the `MPI_Comm_shrink` text based on comments Best, Matthew Whitlock

1 0

Canceled: MPI Fault Tolerance WG Meetings
by Whitlock, Matthew Joseph 02 Jul '26

02 Jul '26

Hello all, We're going to cancel this meeting (7/2) in anticipation of low turnout with the upcoming holiday weekend. Best, Matthew

1 0

FT-WG meeting today
by Joseph Schuchart 18 Jun '26

18 Jun '26

1 0

MPI Fault Tolerance WG Meetings
by Whitlock, Matthew Joseph 18 May '26

18 May '26

Hello all, Updating the time of our meetings, accessed at the Zoom link below. Best, Matthew Whitlock Joseph Schuchart is inviting you to a scheduled Zoom meeting. Join Zoom Meeting https://urldefense.us/v3/__https://stonybrook.zoom.us/j/99412608922?pwd=xAQ… Meeting ID: 994 1260 8922 Passcode: 498354 --- One tap mobile +13097403221,,99412608922# US +12122258997,,99412608922# US Join instructions https://urldefense.us/v3/__https://stonybrook.zoom.us/meetings/99412608922/…

1 0

Canceled: MPI Fault Tolerance WG Meetings
by Whitlock, Matthew Joseph 18 May '26

18 May '26

Hello all, Updating the time of our meetings, accessed at the Zoom link below. Best, Matthew Whitlock Joseph Schuchart is inviting you to a scheduled Zoom meeting. Join Zoom Meeting https://urldefense.us/v3/__https://stonybrook.zoom.us/j/99412608922?pwd=xAQ… Meeting ID: 994 1260 8922 Passcode: 498354 --- One tap mobile +13097403221,,99412608922# US +12122258997,,99412608922# US Join instructions https://urldefense.us/v3/__https://stonybrook.zoom.us/meetings/99412608922/…

1 0

Re: [mpiwg-ft] Fault Tolerance WG Meeting Reminder
by Joseph Schuchart 08 May '26

08 May '26

1 0

Re: [mpiwg-ft] Fault Tolerance WG Meeting Reminder
by Skjellum, Anthony 07 May '26

07 May '26

Hi, is there any chance to shift this meeting an hour earlier or later each week? Thanks, Tony Anthony Skjellum, PhD Professor of Computer Science Director, Advanced Scalable Computing, Extreme Networks & Data (ASCEND) Center Tennessee Technological University email: askjellum(a)tntech.edu cell: +1-205-807-4968 ________________________________ From: mpiwg-ft <mpiwg-ft-bounces(a)lists.mpi-forum.org> on behalf of Whitlock, Matthew Joseph via mpiwg-ft <mpiwg-ft(a)lists.mpi-forum.org> Sent: Thursday, May 7, 2026 11:12 AM To: Aurelien Bouteiller via mpiwg-ft <mpiwg-ft(a)lists.mpi-forum.org> Cc: Whitlock, Matthew Joseph <mwhitlo(a)sandia.gov> Subject: [mpiwg-ft] Fault Tolerance WG Meeting Reminder External Email Warning This email originated from outside the university. Please use caution when opening attachments, clicking links, or responding to requests. ________________________________ Hello all, We have our fault tolerance meeting today at 3pm Eastern. Looking forward to it, Matthew Whitlock ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ‍ ZjQcmQRYFpfptBannerStart This Message Is From an External Sender This message came from outside your organization. ZjQcmQRYFpfptBannerEnd Hello all, We have our fault tolerance meeting today at 3pm Eastern. Looking forward to it, Matthew Whitlock

1 0

Fault Tolerance WG Meeting Reminder
by Whitlock, Matthew Joseph 07 May '26

07 May '26

Hello all, We have our fault tolerance meeting today at 3pm Eastern. Looking forward to it, Matthew Whitlock

1 0

Issues with MPI_Comm_agree output num_agreed_failed
by Whitlock, Matthew Joseph 01 May '26

01 May '26

Hello all, The changes we've been suggesting for MPI_Comm_agree aren't quite as straightforward as we've been thinking. Change 1 is to output a num_agreed_failed such that the first num_agreed_failed of the groups returned by a subsequent call to MPI_Comm_get_failed will be similar across all ranks. However, we cannot guarantee the similarity of the fail groups unless new_failed is false. Consider: MPI_Comm_iagree(..., &num_agreed_failed, &new_failed, &request); MPI_Test(&request, ...); // This rank passes known failures upwards in agreement's reduce step, but bcast step is not complete MPI_Recv(...); // this operation fails, local rank now knows of a new failed rank X MPI_Wait(&request, ...); // agree completes with new_failed. Rank X is not counted in num_agreed_failed if it contributed to the agreement before failing, but it may be placed before any newly detected failures in the local failed group. Change 2 is to allow num_known_failed to be renamed max_num_expected_failed and be potentially larger than the number of locally known failures. All ranks which pass a value larger than the local number of known failed ranks, however, run into the above problem even if new_failed is false. So we could only say that the fail groups up to num_agreed_failed are guaranteed similar if new_failed is false and the passed max_num_expected_failed was <= the size of the local fail group on all ranks. That would be pretty awkward in the standard, so the more realistic solutions are to 1. internally repeat the agreement until !new_failed (performance implications, does not resolve the problem with change 2), 2. return a consistent agreed_failed group instead of num_agreed_failed (extra group allocation, potentially extra data in agreement bcast step, but both could be avoided with an MPI_GROUP_IGNORE), or 3. revert to the original function syntax. I prefer B, which lets you safely optimistically proceed with a consistent returned group of failed ranks instead of calling MPI_Comm_agree in a loop until !new_failed. Implementations can pass a flag indicating if the local rank called with MPI_GROUP_IGNORE during the reduce step to avoid passing the failed ranks in the bcast step (if they even have a way to avoid that in current implementations). Please consider and let me know your thoughts here or at the next meeting. Thanks, Matthew Whitlock

1 0

MPI Fault Tolerance WG Meetings
by Whitlock, Matthew Joseph 23 Apr '26

23 Apr '26

Hello all, Reminder that our next fault tolerance meeting is today, using the Zoom link below. See you there, Matthew Whitlock Joseph Schuchart is inviting you to a scheduled Zoom meeting. Join Zoom Meeting https://urldefense.us/v3/__https://stonybrook.zoom.us/j/99412608922?pwd=xAQ… Meeting ID: 994 1260 8922 Passcode: 498354 --- One tap mobile +13097403221,,99412608922# US +12122258997,,99412608922# US Join instructions https://urldefense.us/v3/__https://stonybrook.zoom.us/meetings/99412608922/…

1 0