Abstract: "The aggregate computing power of a network of workstations is substantial and comparable to that of mainframe computers. We propose a multiprocessing methodology to utilize this computing power effectively in executing compute-intensive applications. In our methodology, an applicatio...

Ausführliche Beschreibung

Bibliographische Detailangaben

Format:	Buch
Sprache:	English
Veröffentlicht:	Menlo Park, Calif. 1989
Schriftenreihe:	Computer Science Laboratory <Menlo Park, Calif.>: SRI-CSL 89,2
Schlagworte:	Computer networks Multiprocessors
Zusammenfassung:	Abstract: "The aggregate computing power of a network of workstations is substantial and comparable to that of mainframe computers. We propose a multiprocessing methodology to utilize this computing power effectively in executing compute-intensive applications. In our methodology, an application is composed of a relatively small number of coarse-grained function modules expressed as conventional, sequential code. These modules are linked together using a directed-graph language (DGL) that allows specification of inter-module data dependencies. The execution system manages the parallel execution of function modules on multiple workstations using demand-driven execution We have implemented a prototype demonstration of the methodology on top of Sun OS running on Sun workstations. We have experimented with two applications, namely, matrix multiplication and the UNIX make program. Results from these experiments suggest that improved performance can be obtained from these applications as more workstations are make available to them. The scalability is limited by the inherent parallelism in the application and by the overhead of the particular implementation.
Beschreibung:	II, 21 S.