An adaptive communications protocol for network computers (extended abstract)

A network computer is a collection of computers designed to function as one machine. On a network computer, as opposed to a multiprocessor, constituent subcomputers are memory-disjoint and communicate only by some form of message exchange. Ensemble architectures like multiprocessors and network comp...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Performance evaluation review 1985-08, Vol.13 (2), p.4-5
Hauptverfasser: Gelernter, David, Podar, Sunil, Badr, Hussein G.
Format: Artikel
Sprache:eng
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:A network computer is a collection of computers designed to function as one machine. On a network computer, as opposed to a multiprocessor, constituent subcomputers are memory-disjoint and communicate only by some form of message exchange. Ensemble architectures like multiprocessors and network computers are of growing interest because of their capacity to support parallel programs, where a parallel program is one that is made up of many simultaneously-active, communicating processes. Parallel programs should, on an appropriate architecture, run faster than sequential programs, and, indeed, good speed-ups have been reported in parallel programming experiments in several domains, amongst which are AI, numerical problems, and system simulation. Our interest lies in network computers, particularly ones that range in size from several hundred nodes to several thousand. Network computers may be organized in either of two basic ways: their nodes may communicate over a shared bus (or series of buses), as in S/Net; or over point-to-point links, as in Cosmic Cube and the Transputer Network. The work to be presented deals with the point-to-point class, the elements of which we shall refer to as “linked networks”. Linked networks face a fundamental communication problem. Unless they are completely connected (which is rarely possible), two communicating nodes will not necessarily be connected by a single link. Messages between nodes must therefore, in general, travel over several links and be processed by several intermediate nodes. Communication delays increase with the length of the traveled path. Network computer designers therefore provide networks the diameters of which are small relative to their size, and network operating systems will attempt to place communicating processes as close to each other as possible. We present a communication protocol for linked networks that was designed specifically for network computers. Staged Circuit Switching is a communication protocol that combines aspects of store-and-forwarding with aspects of circuit switching, where circuit switching refers to the class of protocols in which a communicating source and destination first construct a dedicated path or circuit between them, then communicate directly over this path. The path may be a physical connection, as in spaced-switched circuit-switching, or a series of dedicated slots in time-division multiplexing switches, as in time-switching protocols. The stage-circuit-switching de
ISSN:0163-5999
DOI:10.1145/317786.317803