Petra Schilhard

Distribution of Imunes System: Graph Partitioning Petra Schilhard

In my thesis, I have implemented a graph partitioning heuristic in IMUNES system. IMUNES is a program for network emulation/simulation. Since network emulation and simulation are computationally very intensive, a method for dividing the computational load between distributed processors, in a way that minimizes interprocessor communication, is required. The emulation in IMUNES is represented as graph nodes interconnected with links, so the problem of dividing the computational load between processors is reduced to dividing the nodes and edges of the graph, called graph partitioning. After METIS graph partitioning algorithm, I implemented the graph partitioning which has three stages. In the first stage, coarsening, a hierarchy of approximations to the original problem is created. In the second stage an initial solution to the problem is found, which is then iteratively refined in the third stage. In the main part of this thesis I describe the implementation of the three stages of the multilevel partitioning scheme in IMUNES. The main goal was to implement a real-time job partitioning in IMUNES system.