Study on Parallel Processing of ab initio Crystal Orbital Calculations Using PC Cluster

Abstract

We study the parallel processing environment using the MPI/LAM message passenger. We perform ab initio crystal orbital calculations of one-dimensional polymers using parallel processing. The personal computer cluster of eight dual CPU motherboards with Intel Pentium III 450MHz (total 16 CPU) is connected to a 100BaseT ethernet switch. The calculation on poly-tetrafluoro-ethylene (C₂F₄)_x shows that the wall clock by 8 CPU is 6.27 times faster than that by the single CPU. The wall clock by 16 CPU, however, remains just 6.96 times faster than that by single CPU, which should be attributed to the inefficiency of the dual CPU motherboard on the I/O wait time and the network wait time. The calculation on poly-(para-phenylene sulfide) (C₆H₄SC₆H₄S)_x shows that the wall clock by 8 CPU becomes about 20 times faster than that by the single CPU. This is enabled by the fact that the total system memory becomes greater than the temporary file of two electron integrals, which eliminates the access to the hard disk and I/O wait time. These results show the parallel processing of the electronic structure calculations on the one-dimensional polymers to be very effective in terms of the wall clock.