Performance of Multicore Systems on Parallel Data Clustering with Deterministic Annealing

We present a performance analysis of a scalable parallel data clustering algorithm with deterministic annealing for multicore systems that compares MPI and a new C# messaging runtime library CCR (Concurrency and Coordination Runtime) with Windows and Linux and using both threads and processes. We investigate effects of memory bandwidth and fluctuations of run times of loosely synchronized threads. We give results on message latency and bandwidth for two processor multicore systems based on AMD and Intel architectures with a total of four and eight cores. We compare our C# results with C using MPICH2 and Nemesis and Java with both mpiJava and MPJ Express. We show initial speedup results from Geographical Information Systems and Cheminformatics clustering problems. We abstract the key features of the algorithm and multicore systems that lead to the observed scalable parallel performance.