Performance analysis of Particle in Cell Electromagnetic code using Infiniband Interconnect

Particle in Cell (PIC) codes are widely used in the simulation of many plasma related systems (E.g.: Laser plasma interactions, high power microwave sources). A detailed simulation of these systems requires parallel computing facility with faster CPUs with efficient interconnects. We have setup a 33 node Xeon (Dual socket and Dual core) cluster with double data rate Infiniband interconnect in addition to our existing 144 node Pentium-4 cluster with gigabit interconnect. The basic network performance parameters (latency and bandwidth) and the PIC code performance in 32 node Xeon cluster with double data rate Infiniband interconnect has been studied. The studies carried out using 2 cores per node and 4 cores per node are reported.