Analysis of cluster interconnection network topologies

Cluster computing is an economical alternative for the need of high performance computing, whereas previously was only achieved by parallel supercomputers. Cluster of Workstations (COWS) are built with standard components and interconnected by various interconnection topologies. These interconnection topologies allow for the interprocessor communication to occur. A study has been performed to evaluate the computing power in Giga Floating point Operations per second (GFLOPs) of a cluster of computers consisting of 9 nodes interconnected in star, Channel Bonding, and Flat Neighborhood Network topologies. For this task two applications were developed; one performs a Distributed Matrix Multiplication on all the nodes, the second application performs a Data Transfer between two nodes and measures the round trip time of the transfer. In addition the High Performance Unpack (HPL) benchmark was utilized. Results show that 2-way channel bonding is by far the best alternative with peak performance of 12 GFLOPs, whereas 3-way channel bonding achieved a 14 GFLOPs performance but at a much higher cost, making it not desirable. The Flat Neighborhood Network probes to be effective, but it has a very high administrative cost