F. Piccoli, M. Printista (Argentina), and C. Rodriguez León (Spain)
Division Function, Dynamic Polytope, Hypercubic Com munication, Programming Model.
Though there have been substantial progress, the current status of parallel computing is still inmature. No single model of parallel computation has yet come to dominate developments in parallel computing in the way that the von Neumann model has dominated sequential computing. The design of parallel programs requires fancy so lutions that are not present in sequential programming. There are specific parallel solutions for specific problems, but the ultimate goal is to find a general methodology at least for some classes of problems. We present a general methodology to derive efficient parallel divide and conquer algorithms for distributed memory machines using asyn chronous groups of processors. The strategy allows the arbitrary division and coherent reunification of processor subsets, easing the opportunities of the underlying software to divide the network into independent sub networks. This way the impact of the sub network traffic in the rest of the network is minimized and consequently performance pre dictability is benefited. This methodology is formally de fined introducing a computing model we have called One Thread Multiple Processor Model (OTMP). Its expressive ness is exemplified through two divide and conquer pro grams.
Important Links:
Go Back
