2.5n-step sorting on n/spl times/n meshes in the presence of o(/spl radic/n) worst-case faults

In this paper we propose the robust algorithm-configured emulation (RACE) scheme for the design of simple and efficient robust algorithms that can run on faulty mesh-connected computers. We show that 1-1 sorting (1 key per healthy processor) can be performed in 2.5n+o(n) communication steps and 2n+o(n) comparison steps on an n/spl times/n mesh with an arbitrary pattern of o(/spl radic/n) faults. This running time has exactly the same leading constant as the best known algorithms for 1-1 sorting on an n/spl times/n fault-free mesh. We also formulate the mesh robustness theorem, which leads to a variety of efficient robust algorithms on faulty meshes.