A new horizon for sorting on mesh architectures

The author introduces the use of data duplication in massively parallel architectures as a tool for improving the running time of the basic data movement operations. He demonstrates its use by presenting two algorithms for sorting N items on mesh architectures of √N×√N processors. The first algorithm has an O(N^1/3 log N) running time and requires the use of O(N^2/3) memory locations per processor. The second has an O((N log N)^1/3) running time and requires the use of O(N^2/3/log^1/3N) memory locations per processor and multiple broadcasting buses