A randomized parallel branch-and-bound algorithm

Serial processing and parallel processing have algorithms whose complexity increases slowly with problem size. An algorithm that requires frequent interprocessor communication performs poorly as processors are added. Randomized algorithms employ randomness in their solution method and offer benefits related to speedup and complexity. The calculations require less elapsed time, and the code for decomposing the problem into concurrent processes is much simpler. A randomized algorithm operating on a fixed input takes any one of several possible computation paths, depending upon an internally controlled random process derived from the problem structure. A randomized branch-and-bound algorithm is presented to predict the performance that can be realized from solving large branch-and-bound problems on multiprocessors. Such a computation can survive failures of either processors or memory.