Parallel Algorithms: Theory, Practice and Education

Abstract

Each new computing platform required software developers to analyze the algorithms over and over, each time having to answer the same two questions. Does the algorithm possess the necessary properties to meet the architectural requirements? How can the algorithm be converted so that the necessary properties can be easily reflected in parallel programs? Changes in computer architecture do not change algorithms, but this analysis had to be performed again and again when a program was ported from one generation of computers to another, largely repeating the work that had been done previously. Is it possible to do the analysis “once and for all,” describing all of the key properties of an algorithm so that all of the necessary information can be gleaned from this description any time a new architecture appears? As simple as the question sounds, answering it raises a series of other non-trivial questions. Moreover, creating a complete description of an algorithm is not a challenge, it is a large series of challenges, and some of them are discussed in the paper.