It may be hard for tools or existing analysis techniques to provide assistance to users when this mode is selected. However, there is adequate support through the MPI interface to detect when this mode has been selected and restrict the tool's behavior -- i.e. a tool may have to ignore traffic on communicators where this info key is active. As Jeff pointed out, tools can even prevent the application from providing this info key to the library. I don't really see any necessary fix to the proposal. We could add an advice to users to remind them that they should ensure tools are compatible with the info keys. And the reverse advice to tools writers that they should check info keys for compatibility. ~Jim. On Mon, Dec 14, 2015 at 10:59 AM, Daniel Holmes <[email protected]> wrote:

Hi Jeff,

... tools can just intercept the info operations and strip this key if they can't support it.

Agreed: this should have been response 4 on my list (Markus and I discussed that option too).

Cheers, Dan.

On 14/12/2015 15:54, Jeff Hammond wrote:

Remember that one of the use cases of allow_overtaking is applications that have exact matching, in which case allow_overtaking is a way of turning off a feature that isn't used, in order to get a high-performing message queue implementation. In the exact matching case, tools will have no problem matching up sends and recvs.

If tools cannot handle MPI_THREAD_MULTIPLE already, then I don't really care if they can't support this assertion either.

And in any case, such tools can just intercept the info operations and strip this key if they can't support it.

Jeff

On Mon, Dec 14, 2015 at 7:44 AM, Daniel Holmes <[email protected]> wrote:

...

Hi all,

Markus has reminded me that we discussed the effect of the mpi_assert_allow_overtaking assertion on tools (see below).

The basic problem is that message matching would be non-deterministic and it would be impossible for a tool to show the user which receive operation satisfied which send operation without internally using some sort of sequence number for each send/receive operation. As Markus mentions, this is a similar problem to the MPI_THREAD_MULTIPLE "logically concurrent" issue, with which tools already struggle.

My responses were: 1) the user asked for this behaviour so the tool could simply gracefully give up the linking function and just state the information it knows 2) the tool could hook all send and receive operations and piggy-back a sequence number into the message header 3) the tool could hook all send and receive operations and serialise them to prevent overtaking I don't know the viability/acceptability of any of these approaches.

Preferably before the formal reading at the next face-to-face meeting, we should discuss the possibility that this assertion will disrupt tools and ascertain if we can/want to do anything to mitigate that disruption.

Cheers, Dan.

-------- Forwarded Message -------- Subject: MPI P2P "messages may overtake" assertion Date: Mon, 14 Dec 2015 16:24:44 +0100 From: Markus Geimer <[email protected]> <[email protected]> To: [email protected] CC: Marc-André Hermanns <[email protected]> <[email protected]>

Hi Daniel,

Do you remember our conversation after the MPI BoF at SC'15 about the proposed "messages may overtake" assertion and that it may cause severe problems for tools, as correct message matching is very important for many of them? In fact, the issue here is similar to properly supporting MPI_THREAD_MULTIPLE, which most tools don't do either right now...

Meanwhile I talked to Marc-Andre Hermanns from the Tools WG -- which kick-started a discussion on this end. But I guess this topic also requires some involvement of the P2P WG. For example, it would be good if you could summarize the ideas you had in mind how tools could handle this assertion. Thanks!

That said, I don't think there is any urgency for discussions, as people are slowly leaving for the Christmas break (e.g., I'm already out of office). But we should definitely follow up on this early next year.

Merry Christmas, Markus

-- Dr. Markus Geimer Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH 52425 Juelich, Germany

Phone: +49-2461-61-1773 Fax: +49-2461-61-6656 E-Mail: [email protected] WWW: http://www.fz-juelich.de/jsc

------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------ Forschungszentrum Juelich GmbH 52425 Juelich Sitz der Gesellschaft: Juelich Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender), Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, Prof. Dr. Sebastian M. Schmidt ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------

-- Dan Holmes Applications Consultant in HPC Research EPCC, The University of Edinburgh James Clerk Maxwell Building The Kings Buildings Peter Guthrie Tait Road Edinburgh EH9 3FD T: +44(0)131 651 3465 E: [email protected]

*Please consider the environment before printing this email.*

The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

_______________________________________________ mpiwg-p2p mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-p2p

-- Jeff Hammond [email protected] http://jeffhammond.github.io/

-- Dan Holmes Applications Consultant in HPC Research EPCC, The University of Edinburgh James Clerk Maxwell Building The Kings Buildings Peter Guthrie Tait Road Edinburgh EH9 3FD T: +44(0)131 651 3465 E: [email protected]

*Please consider the environment before printing this email.*

The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

_______________________________________________ mpiwg-p2p mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-p2p