As part of ARSC’s dedication to research
within the high performance computing (HPC) community, the center publishes
the HPC Users’ Newsletter to discuss programming models and techniques,
development tools, applications and other issues of importance to users
of HPC systems. Its focus is on ARSC resources and users.
The HPC Newsletter is a biweekly forum for HPC users worldwide and is
co-edited by Tom Baring and Guy Robinson. Readers submit articles, announce
texts, papers, conferences or talks, or simply discuss their work. Ideas
for articles, questions about high performance computing at ARSC, or other
comments for consideration by the editors can be sent to hpc_users@arsc.edu.
The newsletter began as the weekly T3D Newsletter in 1994, became the
bi-weekly T3E Newsletter with issue 115, and evolved into the ARSC HPC
Users’ Newsletter with issue 201 to support changes within ARSC
and the broader HPC community. As of October 2001, the editors have produced
232 issues. The newsletter is now emailed directly to 420 recipients around
the world.
HPC Newsletter Editors Guy Robinson (left) and Tom Baring (right).
Creative HPC Solutions
There are no textbooks specifically written for rapidly evolving supercomputer
technologies. The ARSC HPC Users’ Newsletter fills this niche as
an information source for solutions to the problems and challenges of
high performance computing. Newsletter contributions address issues such
as tracking performance and technology improvements in the Cray SV1, SV1e,
SV1ex upgrade sequence, and monitoring the differences between the IBM
SP and Cray T3E MPP environments. The frequent publication schedule permits
the rapid sharing of ideas between HPC users and programmers.
Here’s a quick survey of major technical articles
that have appeared over the last year:
Don Morton, of the University of Montana, contributed a two-part
series on using MPI Cartesian Topologies. This included a sample application
that runs on a linux cluster or the T3E, and solves a time-dependent,
heat-diffusion equation using finite difference methods.
The editors introduced users to performance monitoring, using hpm
on the Cray SV1, and perfex on the SGI Origin 3800.
l Brad Chamberlain, of the University of Washington, delivered a real-world
example of improving parallel I/O on the T3E.
The editors demonstrated exactly how an ARSC user could force segldr
to load Cray-optimized scilib routines, rather than the hand-coded
alternatives included in the code for portability.
The editors provided OpenMP code for improving the multitasking performance
of certain Cray FFT library routines.
The editors explained Cray’s interleaved memory architecture,
its benefits and the subtle ways a program can experience memory contention.
Monte Carlo methods rely on random sequences. Psuedo-random vectors
(left) are traditionally used, but low-discrepancy Halton vectors (right)
have several advantages. Issues 136 through 138 discuss these methods.
Quick Tip Q&A
“Quick Tips” appear in every issue. Can you answer some of
the questions from this past year? (Go to www.arsc.edu for answers.)
How can I match MPI receives to sends when the number of messages
isn’t known in advance?
Should I use OpenMP SINGLE or MASTER?
Can I tell if my code is using the SV1 cache effectively?
I know I should eliminate implicit typing from my Fortran code. Is
there an easy way?
Should I eliminate WHERE?
State and National Resource…
The Arctic Region Supercomputing Center supports high performance computational
research in science and engineering with an emphasis on high latitudes
and the Arctic.
The center provides high performance computational, visualization, networking
and data storage resources for researchers within the University of Alaska,
other academic institutions, the Department of Defense and other government
agencies. ARSC is located on the UAF main campus in Fairbanks, Alaska.
Arctic Region Supercomputing Center
| PO Box 756020, Fairbanks, AK 99775 | voice: 907-450-8600 | email:
