Replicated Software Components for Improved Performance

In recent years, CPU evolution has shifted from the continuous increase of speed to an increase in the number of processing cores. In this paper, we propose to take advantage of the new multicore systems, by diverse replication of software components. This approach, complementary to other directions that are being tackled, attempts to obtain a better performance for a macro-component consisting of several diverse implementations of the same specification, by returning, for each operation, the result from the replica that is faster for that operation. We present an early design of a system that implements this approach.