From: Marc Snir on 01-13-97 12:36 PM
(Embedded
image moved dinucci @ nas.nasa.gov
to file: 01/09/97 04:50 PM
PIC32460.PCX)
To: fdinucci@nas.nasa.com
cc: mpi-comment @ mcs.anl.gov, mpi-core @ mcs.anl.gov (bcc: Marc
Snir/Watson/IBM Research)
Subject: Re: unnecessary complexities (was: Public Defined)
--0__=xWJBzENSroYfOJ770eZruaQN4d4spAN8cvpDi8EljOuY6BYdSyFI3Vxm
You basically argue that MPI1 was not designed correctly, and that MPI2
compounds the crime. MPI1 had heated discussions on some of the issues
you mention -- in particular communicators vs tags. The decision went
one way -- the main arguments were that this way leaves more freedom to
implementors -- in particular it allows for different divisions between
hardware/firmware supported functions and software supported functions,
and that this design is safer. The decisions may have been wrong, and
CDS1 may be a superior design. But this is clearly not a fruitful debate
for the MPI forum. Clearly, the MPI forum has to design MPI2 using the
same design principles as MPI1, while other groups or individuals may
push designs built on totally different principles.
*****************************
The primary problem, I believe, is in how the standard is broken down
into components. One-sided communication is taken to be different
from message-passing, rather than a version of it or a component of
it. Communicators (or "hypertags", as I have heard them called) are
taken to be a primitive part of the standard, rather than a
construction on top of tags (so now there are multiple kinds of
communicators and multiple constructors for them *in the standard*).
Types are taken to be opaque objects which are constructed using MPI
operations, rather than just simple arguments (so now there are
multiple kinds of types and multiple constructors for them *in the
standard*). I am not saying that these features should not be
available to users, just that they should not be primitives in the
lowest level of the standard. Even with the current breakdown of
functionality, I have often and vocally described my objections and
suggestions for the one-sided chapter on the mailing list (see the
archives for more info), and was sorry to see the dynamic processes
chapter get so large. Those new "info objects" just open the
floodgates, allowing even different vendors to make things as complex
as they want them to be, apparently because the argument lists
couldn't contain all the complexity desired in the standard.
My most complete answer is to ask you to see CDS1 (pointer in my
previous mail). CDS is not perfect, but it is the level that I was
hoping for, and I expect that all the brain power of the forum could
improve on the idea. It is based on a different philosophy, one where
almost all of the design effort was focussed on keeping the interface
as small and simple as possible. Only after that were some
concessions made to increase efficiency on some architectures. I
think that the decision to base MPI-2 on an unchanged MPI-1 made a
similar philosophy difficult to adopt by the forum.
-Dave
========================================================================
David C. DiNucci | MRJ, Inc., Rsrch Scntst | NASA Ames Rsrch Ctr
dinucci@nas.nasa.gov| NAS (Num. Aerospace Sim.)| M/S T27A-2
(415)604-4430 | Parallel Tools Team | Moffett Field, CA 94035
--0__=xWJBzENSroYfOJ770eZruaQN4d4spAN8cvpDi8EljOuY6BYdSyFI3Vxm
Content-type: application/octet-stream;
name="PIC32460.PCX"
Content-transfer-encoding: base64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--0__=xWJBzENSroYfOJ770eZruaQN4d4spAN8cvpDi8EljOuY6BYdSyFI3Vxm--