itaps-parallel [Fwd: Subcommittee assignments for Parallel Interface Definition]

[Tim Tautges]

[Forwarding for archival]

-------- Original Message --------
Subject: Subcommittee assignments for Parallel Interface Definition
Date: Wed, 03 Oct 2007 14:37:11 -0600
From: Devine, Karen D. <kddevin at sandia.gov>
To: Tim Tautges <tautges at mcs.anl.gov>,   Mark Shephard 
<shephard at scorec.rpi.edu>,   Kenneth Jansen <kjansen at scorec.rpi.edu>, 
Lori A. Diachin <diachin2 at llnl.gov>, Vitus Leung <vjleung at sandia.gov>, 
  Carl Ollivier-Gooch <cfog at mech.ubc.ca>,   Karen Devine 
<kddevin at sandia.gov>, Onkar Sahni <osahni at scorec.rpi.edu>


Honored selectees:

As promised, here are the assignments for draft zero
of the parallel interface.

Our task due by November 16 is to build the interface
for Sections 3.2 (Basic Parallelism) and 3.3 (Partitions)
of the requirements document. There are four categories
of interface needed.

   -  Global queries (including MPI Communicators)
   -  Global IDs
   -  Partition Model
   -  Partitioned Mesh

Here's the plan:
Step 0:  draft what functions are needed in each
          category; due Oct 15.
Step 1:  agree on what functions are needed in each
          category; due Oct. 18.
Step 2:  formalize the syntax for those functions;
          due Oct 25.
Step 3:  review and revise output from Steps 1 and 2,
          and write first draft; due Nov. 16.

Below is an outline of what should be included in each
category.  The requirement numbers are those used in
CVS Version 1.14 of the requirements document,
distributed October 2.

Global Queries:

-  Req-5: MPI Communicators
    We did not agree at the bootcamp whether MPI
    communicators should be set once or passed to each
    function.  It seems natural that an MPI communicator
    be associated with a mesh; that is, a mesh would know
    which processors contained any part of the mesh.
    Different communicators could be passed to services
    like load balancing, but in general, a query like
    "give me the global number of element in this mesh" would
    use the mesh's communicator.  So should we have one
    function that associates a communicator with a mesh?
    Or should both a communicator and a mesh be associated
    with a partition?  Or are there cases where separate
    communicators would be needed for each function?  Or...?

    Also here, we should probably decide what a root set
    really is in parallel.  Should there be one root set
    per processor, with functions invoked on the root set
    returning only info on the processor?  If so, should
    there be an additional layer above the root set
    representing the global mesh?

    Just for my understanding, is a root set equivalent to
    a mesh instance?  If, say, a multiphysics application
    has two meshes, does it have two root sets?

-  Req-6:  Global queries
    Which queries should we support?  Should they all be
    associated with a mesh?

-  This category also enforces requirements
    Req-2, Req-3, and Req-4.

Global IDs:

-  Req-7:  Global IDs
    We agreed at bootcamp that global IDs would be
    represented as an array of integers.  We need to
    define the data types and functions on global IDs,
    as well as their immutability.

Partition Model

-  We decided to have a separate iPart handle that
    would be passed to ITAPS functions.  We'll also
    need some functions that return information about
    a partition.  These functions are required by
    Req-10 to Req-13.

Partitioned Mesh

-  We decided that iPart handles could be passed to
    many ITAPS functions in place of Entity handles.
    We need to now examine whether that decision was
    correct.  So for this section, we need to discuss
    how easy substitution of iPart handles will be and
    what additional capability will be needed.  This
    section deals with Req-14 to Req-18.

OK, now the exciting part -- the part you've all been
waiting for -- your assignments!  We'll divide and
conquer, having each group take care of one category
as follows.

Global Queries:    Argonne (Tim)
Global IDs:        Sandia  (Vitus, Karen)
Partition Model:   RPI     (Mark, Ken, Onkar)
Partitioned Mesh:  UBC     (Carl)

By Oct 15, please complete Step 0 and send the output
to this distribution list for review and discussion.
Thanks.

Karen




-- 
================================================================
"You will keep in perfect peace him whose mind is
  steadfast, because he trusts in you."               Isaiah 26:3

             Tim Tautges            Argonne National Laboratory
         (tautges at mcs.anl.gov)      (telecommuting from UW-Madison)
         phone: (608) 263-8485      1500 Engineering Dr.
           fax: (608) 263-4499      Madison, WI 53706