2000 Volume 6 Issue 2 Pages 75-84
We have performed parallel processing on the ab initio crystal orbital calculations of one-dimensional polymers using the TCGMSG and MPI/LAM message passenger. An IBM RS6000 cluster (4 CPU), and personal computer cluster with Intel Pentium II 450MHz (2CPU) and Celeron 433MHz (2CPU) were used for the computational environment. We found that the Celeron CPU was less effective than Pentium II in the parallel calculations. Our previous work showed that the distribution of the CPU demand did not work well when 16 nodes were used to calculate the two-electron integrals in parallel. We found that the new algorithm incorporated in the present article improved the distribution. The CPU time increased when using the MPI interface, because there is an overhead to transfer the array in MPI_ALLREDUCE function. We also determined that the real through-put time was strongly dependent on the structure of the network system.