A comparison between HPF and PVM for data parallel algorithms on a cluster of workstations using a high speed network

In the data parallelism model, the data is partitioned and distributed to all the machines. Operations (often similar) are performed on each set of data and information is passed between processes until the problem is solved. The data parallel algorithm we have implemented simulates the Koch neural network to estimate the optical flow with the gradient-based approach proposed by Horn and Schunk.