A new approach to distributed computing in high energy physics
Paul Avery, Chandra Chegireddy, John Brothers, Theodore J. Johnson, Aric Zion · AIP conference proceedings · 1992
We present a new software system, UFMulti II, which provides tools for distributing HEP applications across multiple Unix workstations in order to take advantage of parallel computing. It is designed for high energy physics applications such as event reconstruction, simulation, and physics analysis. We discuss here a particular component of UFMulti called NetQueues (Network Queues) which permits the fast exchange of data between groups of processes. The NetQueue system is optimized for high data transfer rates across Ethernet and FDDI networks and can accommodate both event and non‐event records in a flexible way. The results shown in this talk are based on tests on a workstation farm at the University of Florida. Measurements of CPU and I/O performance during these tests, including some interesting effects caused by the I/O loading of the CPU, are presented.