BibTeX
@MISC{Edinburgh_designprinciples,
author = {Gregory Wilson Edinburgh and Gregory V. Wilson},
title = {Design Principles for Message-Passing Systems},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
At present, message passing is the most popular programming paradigm for asynchronous parallelism. Many "general-purpose" message-passing systems already exist, and many more are being developed. This paper argues that most existing systems are too unstructured to support large-scale applications development, describes some of the criteria a truly general message-passing system must satisfy, and outlines some alternative implementation strategies. 1 Introduction A bewildering variety of parallel programming paradigms have been developed or proposed during the past quarter century, from semaphores and mailboxes to futures, shared logic variables, and concurrent objects [BST89, Bal90]. Most of these have grown up in the operating systems and networking communities, and assume that accessing shared variables is inexpensive [Andr91], or that a reliable, indivisible broadcast mechanism exists [BTK90]. Neither of these assumptions holds on distributed-memory MIMD computers (multicompu...
Keyphrases
message-passing system design principle networking community alternative implementation strategy operating system large-scale application development popular programming paradigm logic variable concurrent object bst89 distributed-memory mimd computer general message-passing system indivisible broadcast mechanism introduction bewildering variety inexpensive andr91 past quarter century many general-purpose message-passing system asynchronous parallelism message passing
