Optimal searching algorithms for parallel-pipelined computers

Abstract

A new class of optimal search algorithms is presented for the classical problem of searching an ordered list. The appropriate algorithm is selected as a function of the order, (s,p), of the target processor which may consist of p parallel processors each of which is an s level pipelined processor. The algorithms are structured according to a new integer series, the W-series. The algorithms include the binary search for processors of order (1,1), the Fibonacci search for processors of order (2,1), and well-known parallel searches for processors of order (1,p). Performance of the algorithms is evaluated. The algorithms are well-suited to computers with interleaved or parallel memory banks.

This work was supported by the National Science Foundation under Grants GJ-35584X and GJ-40584 and by the Joint Services Electronics Program under Contract DAAB-07-72-C-0259.