benchmarking mistake?

From: Ioana Cozmuta (icozmuta_at_arc.nasa.gov)
Date: Sun May 30 2004 - 20:12:52 CDT

Hi NAMD developers and users,

I've built NAMD on - a HP machinewith 980 nodes, - each
node consists of dual Intel 1.5GHz Itanium-2 processors
- the system has distributed memory
- runs a version of Linux based on Red Hat Linux Advanced
Server
- the processor allocation is scheduled with the LSF
resource manager
- instead of mpirun the prun command is used to submit
parallel jobs (however when I built charm I used the
mpi_intel libraries)

I've run both tests for charm and namd successfully.
However my benchmark data look weird (see below), like the
processors would not be connected/communicating between
them (each job would be run on only 1 Processor).

Could this be caused by the fact that I am using prun
instead of mpirun? Could it be that NAMD does not handle
well distributed memory systems?
Could it be that the Benchmarking time calculations is
calling some functions that Linux might not have?

Any ideea/suggestion would be welcomed!

PS. My benchmark system has ~120k atoms, uses a cutoff of
17.5 and PBC (unit cell of ~100A)

16 CPU;s
Info: Benchmark time: 1 CPUs 6.01797 s/step 69.6524
days/ns 895144 kB memory
Info: Benchmark time: 1 CPUs 6.00938 s/step 69.553 days/ns
895144 kB memory
Info: Benchmark time: 1 CPUs 5.88953 s/step 68.1659
days/ns 903472 kB memory
Info: Benchmark time: 1 CPUs 5.88063 s/step 68.0628
days/ns 903472 kB memory
Info: Benchmark time: 1 CPUs 5.8768 s/step 68.0185 days/ns
903472 kB memory
Info: Benchmark time: 1 CPUs 5.86992 s/step 67.9389
days/ns 903472 kB memory
32 CPU's
Info: Benchmark time: 1 CPUs 6.07359 s/step 70.2962
days/ns 895160 kB memory
Info: Benchmark time: 1 CPUs 6.07609 s/step 70.3252
days/ns 895160 kB memory
Info: Benchmark time: 1 CPUs 5.93727 s/step 68.7184
days/ns 895160 kB memory
Info: Benchmark time: 1 CPUs 5.94625 s/step 68.8223
days/ns 895160 kB memory
Info: Benchmark time: 1 CPUs 5.92906 s/step 68.6234
days/ns 903488 kB memory
Info: Benchmark time: 1 CPUs 5.9318 s/step 68.6551 days/ns
903488 kB memory
48 CPU's
Info: Benchmark time: 1 CPUs 6.08336 s/step 70.4093
days/ns 895144 kB memory
Info: Benchmark time: 1 CPUs 6.08758 s/step 70.4581
days/ns 895144 kB memory
Info: Benchmark time: 1 CPUs 5.94773 s/step 68.8395
days/ns 895144 kB memory
Info: Benchmark time: 1 CPUs 5.95234 s/step 68.8929
days/ns 895144 kB memory
Info: Benchmark time: 1 CPUs 5.92961 s/step 68.6297
days/ns 911800 kB memory
Info: Benchmark time: 1 CPUs 5.93531 s/step 68.6957
days/ns 911800 kB memory
64 CPU's
Info: Benchmark time: 1 CPUs 6.07016 s/step 70.2564
days/ns 889712 kB memory
Info: Benchmark time: 1 CPUs 6.1032 s/step 70.6389 days/ns
889712 kB memory
Info: Benchmark time: 1 CPUs 5.94602 s/step 68.8196
days/ns 898040 kB memory
Info: Benchmark time: 1 CPUs 5.97289 s/step 69.1307
days/ns 898040 kB memory
Info: Benchmark time: 1 CPUs 5.94148 s/step 68.7672
days/ns 906368 kB memory
Info: Benchmark time: 1 CPUs 5.96578 s/step 69.0484
days/ns 906368 kB memory
96 CPU's
Info: Benchmark time: 1 CPUs 6.04344 s/step 69.9472
days/ns 892448 kB memory
Info: Benchmark time: 1 CPUs 6.05594 s/step 70.0919
days/ns 892448 kB memory
Info: Benchmark time: 1 CPUs 5.91297 s/step 68.4371
days/ns 900776 kB memory
Info: Benchmark time: 1 CPUs 5.92805 s/step 68.6117
days/ns 900776 kB memory
Info: Benchmark time: 1 CPUs 5.90109 s/step 68.2997
days/ns 909104 kB memory
Info: Benchmark time: 1 CPUs 5.91242 s/step 68.4308
days/ns 909104 kB memory

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:18:15 CST