|
|
November 6-12, 2004 |
|
|
|
|
David Lawrence Convention Center, Pittsburgh, PA |
|
|
|
|
Our
booth: 2653
|
 |
|
|
|
 |
We
demonstrate that Synergy can effectively compete with MPICH2 in
performance. This demonstration is for a 3000x3000 matrix multiplication
program. (Synergy Source,
MPI source)
|
|
We then show
that we can kill many of the workers (we had five) without affecting the
application to complete). These computers were borrowed from the CIS
Department Lab.
Thanks Tom!.
|
 |
|
 |
We
can calculate the estimated running times a priori. And, we can generate
the parallel code using a Parallel Markup Langauge (PML) resulting
comparable performance!
Our message is that adding an extra smart
communication layer can bring more benefits than the increased overhead.
And, even the overhead can be offset by the flexibility brought by the
new bridging layer..
|
|
Fedora Release 2
was buggy. Our cluster crashed during failover tests. We had to rebuild
the entire cluster on the floor. That was close. |
 |
|
 |
The exhibit floor
was huge. The best companies in the world came. |
|
Our neighbor is University of Southern California. They
show some really cool 3-D Movie animations.If you like the
"Incredible", these are more incredible!
|
 |
|
|
 |
Want to see fast machines? Here is
a 12-CPU machine claiming 2Tflops in such a small footprint. Imaging if
you had thousands of these...
What if one of them crash? Humm...
This is from the Grape Project at University of Tokyo. |
|
|
|
|
Hard drives. Really big hard
drives. |
 |
|
|
|
|
 |
We had free wireless service during the entire conference. This is
the office that provides it. |
|
|
|
|
It was easy to get lost in
Pittsburgh. Too many bridges and tunnels. Especially when it was
raining, and dark... |
 |
|
|
|
|
 |
After the conference was over, we had a huge party that night. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
