Applications of Parallelism to Current Algorithms for Intelligence Analysis

摘要

The purpose of the document is to detail a study of concurrent processing by the Concurrent Processing Subgroup of the U.S. Army Intelligence Center and School (USAMS) task in association with the Advanced Computing Research Facility (ACRF) at Argonne National Laboratory (ANL). The study centered on the effect of different concurrent architectures (hypercube and shared memory) on Intelligence and Electronic Warfare (IEW) algorithm performance. The study examines implementation of a spatial aggregation algorithm on a hypercube and a shared memory machine with special attention given to data partitioning. The difference in implementation of the algorithm are due to data partitioning, data dependence, and communication between processors. Two parallel machines were used: The Cal Tech-JPL Mark II Hypercube, and the Sequent Balance at the Advanced Computing Research Facility at Argonne National Laboratory. The hypercube is a 32 node concurrent processor consisting of 32 identical processors linked by a communications network. The Sequent Balance is a high-performance, general-purpose computer system that uses 2 to 12 National Semi-conductor Series 32000 CPUs in a tightly-coupled multi-processing architecture. This study indicates that task oriented algorithms with a low degree of data interdependence are better suited to a shared memory implementation and data-driven algorithms to a hypercube implementation. Keywords: Parallel processors; Computer architecture; Aggregation. (JHD)