T. Thorsteinsson and S. Rettrup.
Adv. Quant. Chem. 31, 267-282 (1998).

Abstract

The general CI code PEDICI has been parallelized by decomposing the occurring summation over two-electron integrals. The parallelization was formulated in terms of a “master/slave” model, and realized through use of the “PVM” message passing facility. We have aimed at achieving a reasonably simple implementation for use on machines with intermediate numbers of processors. Exploratory test runs on an IBM SP supercomputer (consisting of RS/6000 model P2SC (120 MHz) nodes) show a very satisfactory performance increase with the number of processors used, as well as encouraging balancing of the workload. Our largest 32-processor test case gives a speed-up factor of 30.27.