Evaluating graph theoretic clustering algorithms for reliable multicasting

In reliable multicast protocols, each data packet being sent must be acknowledged. Collecting the acknowledgments centrally at the sources can cause ACK-implosion and can result in poor scalability. To overcome this, clustering algorithms which use virtual structures to gather acknowledgments were proposed. In this work, we analyze the complexities of three such clustering algorithms: Lorax, k-degree, and Self-adjust. We compare the quality of the virtual structures produced by these! algorithms, focusing on the number of clusters, cluster size, cluster radius, and the optimal positioning of cluster leaders. Our simulation showed that the virtual structure produced by Self-adjust is better in terms of cluster radius and the location of cluster leaders. However, due to the self-adjusting nature of the algorithm, it might take longer time to compute than the other two algorithms.