CHARMM Development Project Forums User Discussion & Questions Installation and Testing Parallel computations on PC clusters

Dear all,
I have a new question about CHARMM:) There are clusters of 32 PCs, each with 2 processors, in our institute. However, the clusters have no complex or expensive communication schemes. I wanna know if I could parallel the computations using CHARMM in an efficient way. And what algorithms will enhance the possibility of parallization!
Thanks,
Sheng

The answer depends a bit on the processor speed and how much you can change about the system software; I am assuming Linux. If the processors are faster than 1 GHz and the cluster is running a Linux 2.4.x kernel, CHARMM may not scale well past 2 processors. Processor speed has increased more rapidly than TCP/IP over ethernet performance, which tends to diminish parallel scaling.

The MPI version affects performance slightly; LAM-MPI seems to scale a little bit better than MPICH.

Reverting to the Linux 2.2.20 kernel and applying the 'tpcfix' patch might also improve performance This worked fairly well for machines in the 400-800 MHz range.

The processors are 2.6GHz, and the cluster is running a Linux 2.4.x kernel. The MPI version we use is MPICH.

It is said that MTS algorithm will improve the parellel performance. Could anyone give me some exceptive advice on algorithms.