Programming with active data
In this paper we have defined a virtual systems architecture (VSA), which combines both data parallel and process parallel models of computation. Data manipulations strategies for the efficient implementation of the VSA have been considered for various data types and algorithms, which show that a packet based virtual array implementation is required. We introduce a packet routing scheme the mad postman that is shown to have considerably better latency properties than previously published results and considered synchronous cellula automata to implement this scheme. In general, the mad postman is most advantageous to use in lower dimension networks, for example 2-D meshes, where latency is usually higher due to higher network diameter. In such networks the efficiency of the mad postman routing will be higher due to the smaller number of dead address digits generated per each packet. Such networks match form to function, and in current implementation technologies (i.e VLSI chips and PCBs), they will therefore provide cheaper hardware, thus allowing for a higher cost/performance ratio to be achieved.
KeywordsVirtual Network Systolic Array Virtual Channel Physical Node Virtual Processor
Unable to display preview. Download preview PDF.
- Reddaway, S. F.: DAP — a distributed array processor, First Ann. Symp. on Computer Architecture, Gainsville, (1973), 61–65.Google Scholar
- INMOS Ltd,: The transputer family 1987, INMOS Ltd, 1987.Google Scholar
- Jesshope, C. R.: Implementations of load balanced active-data models of parallel computation. Concurrent Computations: Algorithms, Architectures and Technology (Eds. S. K. Tewksbury, B. W. Dickinson and S. C. Schwartz) Plenum press, New York.Google Scholar
- INMOS Ltd.: The Ooccam 2 reference manual, Prentice Hall, Hemel Hempstead 1988.Google Scholar
- Hillis, W. D.: The connection machine, MIT Press, Massacvhusetts, 1985.Google Scholar
- Cruz, A. J. O.: The design of a control unit and parallel algorithms for a SIMD computer, Ph.D. Thesis, Southampton University, Dept. of Electronics and Comp. Sci., 1988.Google Scholar
- Yantchev J. and C. R. Jesshope: Deadlock free packet routing with bounded buffering for asynchronous regular arrays of processors, Preprint submitted to IEE Proc E.Google Scholar
- Roscoe, A.W. and Dathi, N.,: The Pursuit of Deadlock Freedom, Oxford University Computing Laboratory technical monograph PRG-57.Google Scholar
- Dally, William J.: A VLSI Architecture for Concurrent Data Structures, Kluwer Academic Publishers, Boston, 1987.Google Scholar