Parallel heuristic search: The autonomous parallel combinatorial search and the scalable parallel combinatorial search engine
Abstract
This thesis proposes new ways of using parallel processing to address the computational problems of combinatorial search. A new parallel heuristic search algorithm is proposed. The autonomous parallel search algorithm utilizes the information accumulated by other processors to help each processor make a better selection of what search space nodes to expand. Using the notation of stepwise more informed heuristic evaluation functions, we prove that the use of consensus reduces the number of nodes explored. The algorithm avoids synchronization overhead since each processor may use the information from the other processors whenever it is available, but may also proceed without it. The algorithm represents a new approach to parallel heuristic search. The scalable parallel search engine is an architecture specialized for solving combinatorial search problems. It combines a modified content addressable memory and scalable search processors. Each scalable search processor is an SIMD processor with one instruction unit and multiple execution units. The configuration of the search processors can be tailored to the search problem at hand to vary the number of search graph nodes explored concurrently and the word width in the nodes. The combination of synchronized operation (within a processor) and asynchronous operation (among the processors) increases the utilization of the resources, overlaps the memory access, and avoids synchronization overhead.
Degree
Ph.D.
Advisors
Jamieson, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.