Cs360-2008

Serial C code candidate for OpenMP incremental parallelization

Link to prose, tarball: [1]

Why? Many for loops that look ripe for parallelization; lots of array traversing.

HPL Benchmarking Assignment

• HPL.dat reference from NetLib - http://www.netlib.org/benchmark/hpl/tuning.html

• To get a time of 4+ mins on tarzan, change line 6 of the default ATLAS HPL.dat to: 600 700 800 900 Ns

HPL, tarzan, 4x1, ATLAS

-	LAM	MPICH	OpenMPI
Run 1	4:50.67	4:46.91	4:35.06
Run 2	4:48.69	4:46.25	4:40.36
Run 3	4:50.32	4:45.00	4:38.26
Avg	4:49.89	4:46.05	4:37.89

HPL BENCHMARK STATISTICS FOR TARZAN CLUSTER

-	2 procs	4 procs
Run 1	68.320s	67.336s
Run 2	67.027s	76.995s
Run 3	66.053s	67.117s
Avg	67.133	70.483s

HPL BENCHMARK STATISTICS FOR TARZAN CLUSTER (4 procs, LAM)

-	Atlas	Goto
Run 1	5m53.306s	6m38.663s
Run 2	5m32.486s	6m24.253s
Run 3	5m32.852s	6m26.475s
Avg	5m39.548s	6m29.797