BencHMAP: benchmark-based, hardware and model-aware partitioning for parallel and distributed network simulation

Abstract

Computer simulation of large-scale and complex networks can be resource intensive. Several tools to parallelize and distribute the simulation to a number of different machines have been developed. One of the main challenges facing users of these tools is how to partition the simulation among the computing resources available. The paper focuses on the development of a framework and methodology (ultimately leading to a semi-automated tool) to partition network simulation. The main distinguishing feature of our approach is that the partitioning is performed in a manner that takes into account the specific distributed computation environment available as well as the specific details of the network model. We derive the relationships between impact factors and the simulation performance from measurements of benchmark experiments. We then apply the derived relations to the given network topology and workload model to construct a weighted graph which we then partition using a graph partitioning tool. Experiments on a 120k-node, 100k-stream network simulation show that the full application of this approach improves the performance of partitioned simulation significantly over other partitioning heuristics.