Parallelism issue

From: Artem Zhmurov (zhmurov_at_gmail.com)
Date: Thu Mar 20 2008 - 15:45:15 CDT

Hi,

We're in the process of testing/reinstalling/configuring NAMD on the
cluster. However, we're faced with several problems:
1. When using charmrun for parallel computing on two nodes (8 cpus
each) the speed of calculation less then when using one node. Network
traffic is huge - all the channel is occupied. CPUs are working ~30%
of the peak productivity. We do have a problem, but there are several
issues that still unclear: If there must be an significant increase in
speed of computation for two nodes in our case? What kind of speed
increase we should wait for? What is the optimal nodes/cpus # in terms
of speed increase while running NAMD on such configuration?
2. When running NAMD using mpirun, in the output log all the data is
repeated by each process. Like all of them simply calculating
everything at the same time instead of working together. Does it mean
that NAMD simply not running in parallel but there are several same
NAMD processes (or NAMD is not configured to run with MPI)?

We have 2 Quad Core Xeon processors on each node, Network is Gigabit Ethernet.

Thank you
Artem

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:20 CST