Re: [mpiwg-rma] RMA WG discussion 12/2014
I don't know what RMA users other than I will expect. But advice to users might be useful to, you know, advise users as to what reasonable expectations are :-) Jeff Sent from my iPhone
On Nov 12, 2014, at 12:59 PM, Jim Dinan <[email protected]> wrote:
So, users won't expect that the interface can remotely complete a small RMA operation such that it overtakes a previously issued large RMA operation?
On Tue, Nov 11, 2014 at 2:02 PM, Jeff Hammond <[email protected]> wrote: It seems only obvious that remote request completion only makes sense for large messages where individual completion makes sense. I can think of numerous contexts where it does, especially in the limit of not being able to register an infinity of pages.
Jeff
On Tue, Nov 11, 2014 at 5:49 PM, Jim Dinan <[email protected]> wrote:
I think the argument against was that can be hard to get fine-grain, per-operation remote completion, and harder still to do it efficiently. So, we could end up with an interface that builds a false expectation where users expect overlap that many implementations can't provide.
~Jim.
On Thu, Nov 6, 2014 at 2:38 AM, Jeff Hammond <[email protected]> wrote:
Barrett didn't have a good argument against these functions except argument count, as I recall. There was some debate as to whether nonblocking flush was better, albeit in an apples-to-oranges way.
I think most people are just frustrated with the grossness of the corner we are painted into with RMA API. Pirate RMA is the natural consequence of past decisions. The other solution to the application's problem is overlapping windows, which would be only mildly awful if not for our inability to support just one memory model.
Jeff
Sent from my iPhone
On Nov 6, 2014, at 6:47 AM, "Underwood, Keith D" <[email protected]> wrote:
So, what, you're telling me I have to start attending again to give you an appropriate amount of difficulty for some of these proposals?
-----Original Message----- From: mpiwg-rma [mailto:[email protected]] On Behalf Of Jeff Hammond Sent: Tuesday, November 04, 2014 12:39 PM To: MPI WG Remote Memory Access working group Subject: Re: [mpiwg-rma] RMA WG discussion 12/2014
Pirate RMA (remote request completion - cannot remember ticket number) needs to be retired if WG is still not in favor. But then again, Barrett is gone :- )
Nonblocking RMA epochs in your SC14 paper should be discussed. That looks promising. Can you create a ticket for it?
Jeff
Sent from my iPhone
On Nov 4, 2014, at 5:48 PM, William Gropp <[email protected]> wrote:
The wiki page already has 397 and I added 460. Note also that there is a list of open tickets on that page; we should try to either adopt or retire the open ones.
Bill
On Nov 4, 2014, at 10:32 AM, Jeff Hammond <[email protected]> wrote:
I would like to discuss https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/397 and the closely related https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/460 in San Jose.
Can we collect the other tickets of interest to people and ask Martin for the appropriate allocation of time?
Thanks,
Jeff
-- Jeff Hammond [email protected] http://jeffhammond.github.io/ _______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
_______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
_______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
-- Jeff Hammond [email protected] http://jeffhammond.github.io/ _______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
_______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
participants (1)
-
Jeff Hammond