The D-RISC—An architecture for use in multiprocessors
This paper explores the relationship between latency, bandwidth, concurrency and CPU performance using both theory and concrete design analysis. The results are applied to uniprocessors, where they suggest a new CPU architecture, and multiprocessor systems, where this architecture is shown to be particularly effective.
A model of asynchronous concurrency using threads which may be created, referenced and read is used to analyse and contrast dataflow and von Neumann CPU design. Dataflow CPUs are constrained by fetch bandwidth, von Neumann CPUs by fetch latency. A simple quantitative theory of VLSI CPU implementation shows that latency and bandwidth constraints may be balanced, increasing performance, by choosing the right concurrency. This motivates the D-RISC architecture, a hybrid of RISC and dataflow machines using dataflow-like thread-switching to hide cache miss latency.
KeywordsMultiprocessor System Concurrent Execution Bandwidth Constraint Dataflow Graph Concurrent Thread
- [Arv86a]Arvind and Culler, D. E. Dataflow Architectures. Annual Reviews Of Computer Science, 1986Google Scholar
- [Arv86b]Arvind and Ianucci, R. A. Two Fundamental Issues in Multiprocessing. Computation Structures Group Memo 226–4.Google Scholar
- [Cla80]Clarke, T. J. W., Gladstone, P. J., Maclean, C. D. and Norman, A. C. SKIM — The S. K. I Reduction Machine Proc. 1980 ACM LISP conferenceGoogle Scholar
- [Cla87]Clarke, T. J. W. Forthcoming Ph. D. Thesis October 1987. Cambridge University.Google Scholar
- [Fur85]Furber, S. The Acorn RISC Machine. Microsystem Design, November 11, 1985.Google Scholar
- [Hoa85]Hoare, C. A. Communicating Sequential Processes Prentice Hall International.Google Scholar
- [Li86]Kai Li and Hudak, P. A New List Compaction Method Software — Practice and Experience Vol 16(2). pp145–163 (Feb 86).Google Scholar
- [Mon86]Monteiro, L. F. and Pereira, F. C. A Sheaf Theoretic Model of Concurrency Proc. Logic in Computer Science Symposium June 16–18 1986 Cambridge, Mass.Google Scholar
- [Mea80]Mead, C. and Conway, L. Introduction to VLSI Systems Addison-Wesley 1980Google Scholar
- [Smi78]Smith, B. J. A Pipelined, Shared Resource, MIMD Computer Proc. 1978 International Conference on Parallel Processing.Google Scholar
- [Sto84]Stoye, W. R., Clarke, T. J. W. and Norman, A. C. Some Practical Methods for Rapid Combinator Reduction Proc 1984 ACM LISP Conference pp 197–204.Google Scholar