Parallel motion estimation on SMP system and cluster of SMPs

Recently, symmetric multiprocessor (SMP) systems have become widely available, both as computational servers and as platforms for high-performance parallel computing. The same trend is found in clusters of SMPs, which are expected to become cost-effective parallel-computing platforms. This paper discusses the implementation of parallel motion estimation algorithms on SMP systems and clusters of SMPs. An MPI programming model is used since it supports shared-memory and distributed-memory architectures. Results show that the parallel ME algorithms scale well for SMP systems (up to 6 processors) and clusters of SMPs (up to 23 processors). Moreover, the performance comparison of various systems consisting of 4 processors is described. In terms of price/performance ratio, it is preferable to have a single node with dual-processors rather than a single-processor.