An Efficient Parallelization of Dynamic Programming on a Ring of Processors

c, for different instances of the problem. We discuss how to implement efficiently (1) on a ring of general-purpose processors (Intel Paragon). Our approach is based on the result in [AR94] which provides an analytic solution of optimal tiling problem, namely the problem of determining the size of iteration space tiles so as to minimize the running time of the program. This solution uses a realistic model of the architecture which accounts for coprocessors that permit overlapping of communication and computation, context switching times, etc. Determining the optimal tile size is shown to reduce to a non-linear optimization problem. The closed form analytic solution involves only parameters of the architecture and program that are easily determined at compile time. In this study we improve and develop the model using the peculiarities of Intel Paragon showing that the main ideas from [AR94] are valid also in this case. Figure 1 gives an example which illustrates the corroborati