List Ranking on a Coarse Grained Multiprocessor

We present a deterministic algorithm for the List Ranking Problem on a Coarse Grained p-Multiprocessor (CGM) that is only a factor of log*(p) away from optimality. This statement holds as well for counting communication rounds where it achieves O(log(p) log*(p)) and for the required communication cost and total computation time where it achieves O(n log*(p)). We report on experimental studies of that algorithm on a variety of platforms that show the validity of the chosen CGM-model, and also show the possible gains and limits of such an algorithm. Finally, we suggest to extend CGM model by the communication blow up to allow better a priori predictions of communication costs of algorithms.