[Mpi3-hybridpm] Better Forum INIT/FINALIZE slides

Jeff Squyres (jsquyres) jsquyres at cisco.com
Mon Jul 29 16:19:48 CDT 2013


Yes, agreed -- I just put that box there because it's an obvious thing we'll all see before we send this to the Forum.  It can be discussed on the call in 2 weeks (but I won't be there), or on the call in 4 weeks (I'll be on that one).

My $0.02: because of this proposal, INITIALIZED/FINALIZED *have* to change:

1. they either have to reflect the epoch, like I proposed
2. or they have to be specifically scoped to only apply to the *first* MPI initialization/finalization (which seems really weird)

I'm guessing you're advocating #2, right?  If so, can you explain why?


On Jul 29, 2013, at 3:10 PM, Pavan Balaji <balaji at mcs.anl.gov> wrote:

> Jeff,
> 
> We need to decide on INITIALIZED/FINALIZED vs. IS_WITHIN_EPOCH in the working group and take the final decision to the Forum, instead of wasting the Forum's time on that.
> 
> -- Pavan
> 
> On 07/29/2013 11:43 AM, Jeff Squyres (jsquyres) wrote:
>> I made a few more edits while Brian/Jim/Keith/Pavan are still arguing.  :-)
>> 
>> Here's a new copy...
>> 
>> 
>> 
>> 
>> 
>> On Jul 29, 2013, at 12:32 PM, Jeff Squyres <jsquyres at cisco.com>
>>  wrote:
>> 
>>> Here's a new copy of the slides; if we have time today, we can discuss.
>>> 
>>> I took all feedback, except I disagreed with Steve's #5 and #8.
>>> 
>>> 
>>> 
>>> On Jul 17, 2013, at 9:22 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
>>> 
>>> > Jeff,
>>> >
>>> > Some comments --
>>> >
>>> > 1. Slide 2: The main goals need to be rephrased.  The goals are to overcome limitations while using MPI in stacked and threaded libraries.  For stacked libraries, the single INIT/FINALIZE limitation requires each library to have global knowledge on who  else is using MPI.  For threaded libraries, there are race conditions.
>>> >
>>> > 2. Slide 5: Sorry, I missed this earlier.  I'm not in favor of changing the current definitions of IS_INITIALIZED and IS_FINALIZED.  Better to deprecate them and add MPI_IS_WITHIN_EPOCH.
>>> >
>>> > 3. Slide 10: I don't think we agreed that the requested level will be ignored by subsequent calls.  That's a quality of implementation issue.
>>> >
>>> > 4. Slide 15: We can say that our current stand is to remove that restriction that the "main thread" needs to finalize.
>>> >
>>> > Btw, I've requested for plenary time.
>>> >
>>> > -- Pavan
>>> >
>>> > On 07/15/2013 12:45 PM, Jeff Squyres (jsquyres) wrote:
>>> >> Here's a better cut of slides that are more Forum-friendly.
>>> >>
>>> >> Comments welcome.
>>> >>
>>> >> (Pavan: I'm correct in thinking that you're getting plenary time from Martin for this, right?)
>>> >>
>>> >>
>>> >>
>>> >> _______________________________________________
>>> >> Mpi3-hybridpm mailing list
>>> >> Mpi3-hybridpm at lists.mpi-forum.org
>>> >>http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
>>> >>
>>> >
>>> > --
>>> > Pavan Balaji
>>> >http://www.mcs.anl.gov/~balaji
>>> 
>>> 
>>> --
>>> Jeff Squyres
>>> jsquyres at cisco.com
>>> For corporate legal information go to:http://www.cisco.com/web/about/doing_business/legal/cri/
>>> 
>>> <init-finalize-issues-for-forum.pptx>
>> 
>> 
>> --
>> Jeff Squyres
>> jsquyres at cisco.com
>> For corporate legal information go to:
>> http://www.cisco.com/web/about/doing_business/legal/cri/
>> 
> 
> -- 
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji


-- 
Jeff Squyres
jsquyres at cisco.com
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/





More information about the mpiwg-hybridpm mailing list