(j3.2006) (SC22WG5.3745) Nick's MPI non-blocking proposal

[Brian Smith]

Sorry Van.  Pushed the wrong button so responded to your email instead of 
the person intended.

----- Original Message ----- 
From: "Van Snyder" <Van.Snyder at jpl.nasa.gov>
To: "fortran standards email list for J3" <j3 at j3-fortran.org>
Sent: Thursday, December 04, 2008 5:09 PM
Subject: Re: (j3.2006) (SC22WG5.3745) Nick's MPI non-blocking proposal


>
> On Thu, 2008-12-04 at 15:14 -0800, Aleksandar Donev wrote:
>> Bill Long wrote:
>>
>> > Actually, I find it massively confusing.
>> No surprise, since you are confused about what Nick means by I/O.
>> Nick's proposal is meant to cover send/receive as well. It is I/O in
>> common computer science lingo, .... Just replace I/O with "data
>> transfer" and you should be less confused reading his draft.
>
> I haven't used MPI for 12 years, so there is much about it I don't
> remember, and I'm sure there are developments since I stopped using it.
> But...  My recollection was that it was an enormous kludge that does
> almost nothing other than I/O.  My present codes use PVM, but only for
> inter-program data transfer.
>
> Yeah, there's some synchronization stuff, too, but if you know you have
> to accomplish this using interprocess or interprogram communication, as
> opposed to inter-image shared-memory communication, you might as well do
> that with I/O, too.
>
> Therefore, Nick's proposal inevitably adds yet another kludge to the
> dung heap.
>
> Several times, it has been proposed to add one more spec to the OPEN
> statement to give extra processor-dependent information to the
> processor, with notes pleading to use the information to set up an
> interprocess communication channel, perhaps using MPI, PVM or some
> proprietary method.  I proposed this in 1986, 97-114:17.3 and 08-204,
> and others have proposed it, too.  08-204 proposed adding one row to a
> syntax rule, a dinky one-paragraph subclause in Clause 9, one item in
> Annex A, and one small paragraph in Annex C -- altogether 19 lines of
> plain text, and inevitably much less in typeset LaTeX.  Compare this to
> the 406 pages of MPI-3.
>
> If that were in place, perhaps with the addition of arrays of unit
> numbers in WRITE statements for broadcasts, and if asynchronous I/O and
> DIO worked with it (and why wouldn't they?), why would anybody use MPI
> or PVM for anything other than legacy codes?  If it had been done in
> Fortran 90, would anybody ever have used MPI or PVM in Fortran?
>
> -- 
> Van Snyder                    |  What fraction of Americans believe
> Van.Snyder at jpl.nasa.gov       |  Wrestling is real and NASA is fake?
> Any alleged opinions are my own and have not been approved or
> disapproved by JPL, CalTech, NASA, the President, or anybody else.
>
> _______________________________________________
> J3 mailing list
> J3 at j3-fortran.org
> http://j3-fortran.org/mailman/listinfo/j3
>