H-PBS: a hash-based scalable technique for parallel bidirectional search

Abstract

W e introduce a very simple and efficient hash-based algorithm f o r parallel bidirectional search. Hashing is used i n a novel way to continuously provide every processor with a good cross section of the search front. This hashing serves multiple purposes of achieving load-balancing, selecting nodes for expansion, eliminating duplicate nodes, and detecting termination. Our asynchronous multiprocessor algorithm is shown to perform remarkably well, with linear or superlinear speedups, and scale well for large problem sizes and massive parallelism. The alsorithm is architecture-independent and may be implemented o n various distributed-memory multiprocessors. T h e performance is fairly independent of the communication latency, and takes advantage of a relatively fast communication throughput.W e verify the performance of the proposed algorithm by using the popular 15-puzzle search problem and implementing it o n an 80-processor distributedm e m o r y multiprocessor, EM-4. W e use hundred randomly generated problem instances of the 15-puzzle, with optimal path-lengths i n the range 40-66. I n spite of the limited memory capacity of this parallel machine, our implementation solves 99 of the problem instances, with solution path-lengths very close to optimal, speed-ups ( i n the number of nodes and execution times) close to linear, and in some cases superlinear, and o n the average in 9-4 seconds.

Fast and Efficient Hashing for Sequence Similarity Search using Substring Extraction in DNA Sequence Databases

“…The initial part called the pre contains the first 4 characters of the substring. The remaining string is divided into string of 4 characters and is stored in the post array [8]. Pre is then passed to the search process for further processing.…”

“…Further, [8] considers repeated values for the processing, which will lead to longer strings. In our process, since redundancies are reduced, the maximum size of strings available will be 4 (since only 4 characters are used for representing DNA sequences).…”

Point‐to‐point and multi‐goal path planning for industrial robots

“…16 In each search iteration, the current best node is selected alternatively from the two lists processing both searches simultaneously. The overall run-time is at most twice the runtime of the fastest version of the forward and backward search processed separately.…”

Multi-directional search with goal switching for robot path planning

“…In the second way, two separate OPEN-lists are used for the forward and the backward search (confer [Nassimi95]). In each search iteration, the currently best node is selected alternatively from the two lists, thereby processing the forward and backward search simultaneously.…”