Hi,

Balaji ([EMAIL PROTECTED]) has tested mom4 in mpi (common usage at
GFDL) and SHMEM.  The answers agree to the bit for a test case with
full physics. Here are some comments of note.

The overall difference in the efficiency is small as mom4 with full
physics load is not limited by communication, but by computation.  The
advantage of SHMEM is to reduce the latency which will become apparent
as the 2D communication (free surface) takes a larger fraction of the
total time. SHMEM will also show a greater advantage at larger PE
counts.

To run mom4 in SHMEM, just turn on -Duse_libSMA instead of
-Duse_libMPI.  Also add the -lsma to LDFLAGS. No other change is
required.

Steve Griffies
begin:vcard 
n:Griffies;Stephen
x-mozilla-html:FALSE
org:DOC/NOAA/OAR/GFDL
adr:;;;;;;
version:2.1
email;internet:[EMAIL PROTECTED]
title:Physical Scientist
x-mozilla-cpt:;-11168
fn:Stephen Griffies
end:vcard

Reply via email to