Academic Commons

Reports

DFLOPS: A Data Flow Machine for Production Systems

Cheng, Fu-Chiung; Wu, Mei-Yi

Many production system machines have been proposed to speed up the execution of production system programs. Most of them are implemented based on conventional control flow model of execution which is limited by the "von Neumann bottleneck." In this paper we propose DFLOPS, a new multiprocessor data flow machine, for parallel processing of production systems. Rule programs are compiled into data flow graphs and then mapped into DFLOPS processing elements. Three levels of parallelism: Rule Level Parallelism, RHS Level Parallelism and LHS Parallelism are fully exploited to achieve high performance. The design and implementation of DFLOPS is presented in detail. The distinguishing characteristics of this proposed machine lies in its simplicity, fully-pipelined processing and fine grain parallelism. The initial results reveal that the performance of production systems is greatly improved.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-025-93
Published Here
January 27, 2012
Academic Commons provides global access to research and scholarship produced at Columbia University, Barnard College, Teachers College, Union Theological Seminary and Jewish Theological Seminary. Academic Commons is managed by the Columbia University Libraries.