MPI - HPC on Linux cluster

The MPI add-on for High-Performace Computing enables GRASP to run on Linux clusters or any collection of Linux computers. This allows a significant reduction of the GRASP computation times which in turn can reduce development time and costs. The MPI add-on is delivered as a separate cluster edition of GRASP and we usually work closely with the end-customer to integrate GRASP in existing MPI runtime environments or batch scheduling environments.

The cluster-enabled version of GRASP features load-balanced computation of currents and fields, using standard GRASP input files. The input files can be prepared using GRASP for Windows or Linux and then executed on the cluster. The output files will be exactly identical to those produced by the serial version of GRASP. The performance of the cluster-enabled GRASP has been tested on a 24-node cluster and the measured speedup is shown in the plot below. For this dual-reflector PO test case, the speedup was approximately 20. Our customers have reported GRASP runs on up to 128 nodes.

The GRASP cluster edition contains 64-bits executables for clusters connected by Ethernet network. Please ask if you are interested in other network fabrics, such as Infiniband or Myrinet.