[Mpi3-hybridpm] External interfaces chapter updates
Bronis R. de Supinski
bronis at llnl.gov
Fri Nov 5 18:52:38 CDT 2010
Sorry I had to drop out of the call early today but
I had to be somewhere at noon.
Anyway, I am looking over Doug's comments now.
Pavan and Doug:
> Doug's first issue indicates that you need to update
> your PDF again. I have fixed that issue (which you
> caught earlier) in the current draft in the MPI-3.0
> trunk. Please start with the version that is now in
> the MPI-3.0-2010-11-draft branch as I have made the
> other minor corrections that we have discussed in
> that section.
Looks good now. Thanks.
> I will look over Doug's other points to see which
> should be included in that draft as well as your
> proposed version once you update it.
> On Mon, 1 Nov 2010, Douglas Miller wrote:
>> Just proof reading again...
>> Page 11 lines 32-34, this paragraph still says "Advice to implementers"
This issue should now be fixed in all versions.
>> Page 12 lines 3-4, missing close-paren at "async-signal-safe".
Good catch. fixing it requires that we move the period outside
of the quotation marks so it can follow the close-paren. I have
made this change and commited it to MPI-3.0-2010-11-draft branch
in which we are storing changes that will not go out in the
first release draft since they have not been discussed/seen by
the overall Forum (I know this seems like it should be OK to
include in that draft but I am just following my understanding
of the rules; it should get into the next release draft).
>> Page 12 lines 46-48, I think MPI_THREAD_FUNNELED should have the same note
>> about excepting helper threads calls? I think it can still be allowed, or
>> might be desirable, to use helper threads in this case?
I do not understand this comment. Which note? This confusion
is probably an issue over which version of Pavan's document
you used for page/line numbers. However, I think this comment
and all remaining ones only pertain to the version with more
significant changes (i.e., the helper threads and shared memory
proposals). I don't intend to integrate them into the branch
with the small changes yet so I will stop here. Please let
me know if I have misinterpreted something.
Where are you keeping the version with the more significant
changes? Did you have Jeff cut you a branch? We should definitely
do something to keep those working changes available. It would
probably be good if both of us had write access to them in
case something happens and you need someone else to pick it up.
>> Page 14 lines 42.5-48, the wording sounds a little soft, as if the only
>> goal is to pass the linking phase. should it additionally say something
>> like "must return a meaningful and accurate value"?
>> Pages 16-17, should we add an "advice to users" to recommend/remind that
>> all communications between JOIN and LEAVE be self-completing? What I mean
>> is that if a thread does an MPI_ISEND between JOIN and LEAVE, that is also
>> does an MPI_WAIT (or equiv.) on that ISEND before LEAVE? Since JOIN and
>> LEAVE have no knowledge of requests, etc, isn't that prudent or even
>> Page 17, section 12.5 Shared Memory. Rather than be collective, could these
>> calls reflect the API of something like shm_open() whereby they have a
>> "key" parameter that uniquely identifies the segment of shared memory? Our
>> experience with DCMF (where we did all shmem allocations in a ordered,
>> synchronized "collective" manner) was that it is fraught with problems and
>> restrictions. We're moving to using an API that takes a string "key" so
>> that we need not force such semantics. Are there any OS shmem APIs that
>> require ordered, collective allocation? I know UPC does not use a "key",
>> but wouldn't this allow for better implementations? Are there platforms
>> where these semantics would NOT work? [probably a topic for our meeting]
>> [also another topic for the meeting] Should we say something about how to
>> get a communicator of appropriate ranks for shmem allocation? Many
>> platforms do not support global shared memory (only shmem local to a node),
>> and I don't think there are any MPI mechanisms for testing or selecting
>> ranks that are node-local.
>> Thanks, Pavan, for doing this integration, by the way.
>> Myself, I don't know LaTeX and can't afford the learning curve right now,
>> so you're really helping out.
>> Douglas Miller BlueGene Messaging Development
>> IBM Corp., Rochester, MN USA Bldg 030-2 A410
>> dougmill at us.ibm.com Douglas Miller/Rochester/IBM
>> [cid:1__=09BBFD5DDFD0140C8f9e8a93df938 at us.ibm.com]Pavan Balaji
>> ---10/30/2010 07:37:24 PM---On 10/30/2010 01:49 PM, Bronis R. de Supinski
>> Pavan Balaji <balaji at mcs.anl.gov>
>> Sent by: mpi3-hybridpm-bounces at lists.mpi-forum.org
>> 10/30/2010 07:36 PM
>> Please respond to
>> mpi3-hybridpm at lists.mpi-forum.org
>> "Bronis R. de Supinski" <bronis at llnl.gov>,
>> mpi3-hybridpm at lists.mpi-forum.org
>> Re: [Mpi3-hybridpm] External interfaces chapter updates
>> On 10/30/2010 01:49 PM, Bronis R. de Supinski wrote:
>>>> I think it's time for us to start dumping in text into the chapter and
>>>> start discussing the exact wording. I've included the helper threads and
>>>> shared memory extensions proposals into the chapter and uploaded it to
>>>> the wiki
>>>> Please take a look at it and let me know your comments.
>>> I'll try to make a detailed reading next week. In the
>> I had initially incorrectly uploaded a change with respect to the
>> MPI_THREAD_SERIALIZED semantics which we decided to drop last time. I've
>> now uploaded v0.2 of the document with this correction as well as the
>> other changes suggested by Bronis.
>> -- Pavan
>> Pavan Balaji
>> Mpi3-hybridpm mailing list
>> Mpi3-hybridpm at lists.mpi-forum.org
More information about the mpiwg-hybridpm