Scalable combinatorial search engine

Abstract

A new parallel architecture for solving combinatorial search problems is introduced. The scalable parallel combinatorial search engine is composed of a modified content addressable memory and combinatorial search processors. A flexible interconnection network allows the system to be tailored to the search problem. The combination of synchronous operation (across the multiple execution units within a processor) and asynchronous operation (among the processors) increases the utilization of resources, overlaps memory access, and avoids synchronization overhead. Analysis shows that speedup is linear in the total number of execution units in the search engine. Also, the worst case space complexity for a problem size of n + 2 states with predetermined beginning and ending states can be reduced to 2/sup n/-1 n//spl Sigma//k = 1 (n/k-1) /spl times/ (k-1) of the complexity of a combinatorial search. When n increases from 10 to 50, this reduction factor decreases from 0(10/sup -4/) to 0(10/sup -50/).