A comparison of MPI and Charm++ parallel programming technologies on the minimum spanning tree problem

The paper presents implementations of the GHS minimum spanning tree algorithm developed using message passing model (MPI library), message-driven model (Charm++ language), and vertex-centric model in Charm++. The optimized GHS implementations using MPI and Charm++ have approximately the same performance on 32-node cluster, the performance degradation of the implementation in Charm++ vertex-centric model is of 1-2 orders of magnitude.