J. R Gurd
Abstract
The Manchester project has developed a powerful dataflow processor based on dynamic tagging. This processor is large enough to tackle realistic applications and exhibits impressive speedup for programs with sufficient parallelism.
- 1 IEEE. Special issue on dataflow systems. ZEEE Comput. 15, 2 (Feb. 1982).Google Scholar
- 2 Gurd, J.R., Watson, 1. Kirkham, CC., and Glauert. J.R.W. The dataflow approach to parallel computation. In Distribufed Computing, F.B. Chambers, D.A. Duce, and G.P. Jones, Eds. APIC Studies in Data Processing. vol. 20, Academic Press, New York, Sept. 1984.Google Scholar
- 3 Glauert, J.R.W. High level languages for dataflow computers. State of the Art Rep. Ser. IO. Number 2, on Programming Technology, Pergaman-Info&h. Maidenhead. U.K. Mar. 1982.Google Scholar
- 4 Treleaven. P.C. Brownbridge, D.R., and Hopkins, R.P. Data-driven and demand-driven computer architecture. ACM Compuf. Sure. 14,l (Mar. 1982). 93-143. Google ScholarDigital Library
- 5 Young. J.W., and Kent, H.K. Abstract formulation of data processing problems. Intern. Rep., Product Specifications Dept., The National Cash Register Company, Hawthorne, Calif., 1958.Google Scholar
- 6 Brown, G.W. A new concept in programming. In Computers and the World of the Future, M. Greenberger, Ed. MIT Press, Cambridge, Mass., 1962.Google Scholar
- 7 Karp. R.M. and Miller, R.E. Properties of a model for parallel computations: Determinacy, termination and queue@ SIAM J. Appt. Math. II, 6 (Nov. 1966), 1390-1411.Google Scholar
- 8 Adams, D.A. A computational model with data flow sequencing. Ph.D. thesis, TR/CS-117, Dept. of Computer Science, Stanford Univ., Calif. 1968. Google ScholarDigital Library
- 9 Rodriguez. J.E. A graph model for parallel computation. Ph.D. thesis, MIT/LCS/TR-64. Laboratory for Computer Science, MIT, Cambridge, Mass., 1969. Google ScholarDigital Library
- 10 Dennis, J.B. Fosseen. J.B., and Linderman. J.P. Data Flow Schemas. Lecture Notes in Computer Science, vol. 5. Springer-Verlag, New York, 1974.Google Scholar
- 11 Dennis. J.B. First Version of a Data Flow Procedure Language. Lecture Notes in Computer Science. vol. 19. Springer-Verlag. New York, 1974. Google ScholarDigital Library
- 12 Dennis, J.B., and Misunas, D.P. A preliminary architecture for a basic data flow architecture. In Proceedings of the 2nd Annual Symposium on Computer Architecture. IEEE Press, New York, Jan. 1975, pp. 126-132. Google ScholarDigital Library
- 13 Syre. J.C., et al. LAU system-A parallel data-driven software/hardware system based on single-assignment. In Parallel Compufers- Parallel Mathematics. M. Feilmeier. Ed. Elsevier North-Holland. New York, 1977.Google Scholar
- 14 Johnson, D. Automatic partitioning of programs in multiprocessor systems. In Proceedings of fhe IEEE COMPCON, IEEE Press, New York, Apr. 1980.Google Scholar
- 15 Miranker. G.S. Implementation of procedures on a class of data flow processors. In Proceedings of the IEEE International Conference on Parallel Processing, IEEE Press, New York, Aug. 1977.Google Scholar
- 16 Davis, A.L. The architecture and system method of DDMl: A recursively structured data driven machine. In Proceedings of the 5th ACM Symposium on Computer Architecture. SIGARCH Newsl. 6. 7 (Apr. 1978), 210-215. Google ScholarDigital Library
- 17 Arvind. Gostelow. K.P., and Plouffe, W. An asynchronous programming language and computing machine. Tech. Rep. TR114a, Dept. of Information and Computer Science, Univ. of California, Irvine, Dec. 1978.Google Scholar
- 18 Gurd, J.R., Watson, I., and Glauert, J.R.W. A multilayered data flow computer architecture. Intern. Rep. Dept. of Computer Science, Univ. of Manchester, England. Jan. 1978.Google Scholar
- 19 Tesler, LG. A language design for concurrent processes. In Proceedings of AFIPS Spring joint Computer Conference {Atlantic City, N.J., Apr. 30-May 2). AFIPS Press, Montvale, N.J., 1968, pp. 403-408.Google ScholarDigital Library
- 20 Chamberlin. D.D. The "single-assignment" approach to parallel processing. In Proceedings of AFIPS Fall Joint Computer Conference (Las Vegas, Nev. Nov. 16-18). AFIPS Press, Montvale, N.J., 1971, pp. 263- 270.Google Scholar
- 21 McGraw, J., et al. SISAL-Streams and iteration in a singleassignment language. Language Reference Manual (version 1.0). Lawrence Livermore National Laboratory, Livermore. Calif. July 1983.Google Scholar
Index Terms
- The Manchester prototype dataflow computer
Recommendations
Iterative Instructions in the Manchester Dataflow Computer
The authors investigate the nature and extent of the benefits and adverse effects of iterative instructions in the prototype Manchester Dataflow Computer. Iterative instructions are shown to be highly beneficial in terms of the number of instructions ...
Dataflow Predication
MICRO 39: Proceedings of the 39th Annual IEEE/ACM International Symposium on MicroarchitecturePredication facilitates high-bandwidth fetch and large static scheduling regions, but has typically been too complex to implement comprehensively in out-of-ordermicroarchitectures. This paper describes dataflow predication, which provides per-...
Scheduled Dataflow: Execution Paradigm, Architecture, and Performance Evaluation
Special issue on the parallel architecture and compilation techniques conferenceIn this paper, the Scheduled Dataflow (SDF) architecture a decoupled memory/execution, multithreaded architecture using nonblocking threads is presented in detail and evaluated against Superscalar architecture. Recent focus in the field of new processor ...
Reviews
Reinhard Wilhelm
The data flow model of computation can be characterized by a graphical representation of programs (i.e., the (directed) data flow graph) and by the firing rule for the nodes of the graph. The nodes representing instructions of the program fire as soon as arguments are available at all incoming arcs. The result of the executed instruction is then passed on to all consuming instructions. Languages implemented on this model may provide (recursive) procedures and loops requiring cycles in the graph or dynamic copying of subgraphs, resp. Hence, different iterations of a loop (incarnations of a procedure) may have to share a subgraph. The Manchester project uses tags on the data tokens to prevent different incarnations of a subgraph from interfering with each other while allowing concurrent computation in a subgraph. The data flow engine built at the University of Manchester, like most data flow computers designed so far, consists of a multiprocessor using message passing on a ring. All data flow graphs resulting from translation of data flow programs must be executable on this fixed architecture. The Manchester engine uses labeled nodes which contain the instruction to be performed and the destination of the result. These nodes are kept in the Instruction Store. When Instruction Store receives a packet containing a node's label and the necessary operands, it generates an executable packet and sends it to the Processing Unit. The Processing Unit executes the instruction and sends the result packet through the Token Queue to the Matching Unit, which groups partner tokens together to send them to the Instruction Store. The paper reports on data obtained by simulation and by experiments made with the prototype. Only a preliminary evaluation is possible based on these data. But weak points in the first design could be identified and corrected.
Access critical reviews of Computing literature here
Become a reviewer for Computing Reviews.
Comments