Performance evaluation of distributed maximum weighted matching algorithms
Can Umut Ileri, Orhan Dağdevıren · 2016
Graph matching is a fundamental graph theory problem which has a broad application range including information retrieval, pattern recognition, graph partitioning, chemical structure analysis, protein function prediction, backup placement and cellular coverage. This problem has gained attention in distributed computing as there are distributed matching algorithms with asymptotically guaranteed time bounds and approximation ratios. On the other side, we do not know the practical performance of these algorithms. In this paper, we provide a detailed performance evaluation of asynchronous distributed maximum weighted matching (MWM) algorithms. We assume a message-passing system in CONGEST model in which the message size is limited to O(log n) where n is the number of nodes. This model is popular for energy-efficient networks such as wireless sensor networks. We used a discrete event simulator, SimPy, to model the assumed network structures. We provide the implementations of Watthenhofer and Wattenhofer's algorithm, Hoepman's algorithm, Lotker et al.'s algorithm and Lotker et al.'s improvement algorithm. The results show that the greedy algorithm of Hoepman performed best in approximating the optimum result in all types of networks, even achieving an approximation ratio of 0.99 in some instances. To the best of our knowledge this is the first study which provides an extensive performance evaluation of distributed MWM algorithms.