Skip to Main Content
Summary form only given, as follows. The efficient implementation of the FRED electromagnetic particle-in-cell (PIC) code on a modern distributed memory supercomputer is made difficult by the conflicting objectives of balancing the computational load and minimizing interprocessor communication. The irregular nature of the Lagrangian particle description in the FRED PIC code results in time-fluctuating data communication patterns which hinder efficient utilization of distributed memory parallel computers because of the resulting overhead for frequent data redistribution and dynamic load balancing. Some possible approaches to the resolution of these issues are presented and experience with one such approach on the LANL Blue Mountain (SGI Origin 2000) computer is discussed. The motivation for distributed memory implementation of the FRED code is to enable practical calculation of the plasma evolution of plasma opening switches in the 10**14 electron number density, /spl sim/100 ns timescale regime for volumes of /spl sim/1000 cc.