Evaluation of gossip to build scalable and reliable multicast protocols

This paper provides an extensive experimental evaluation of the gossip technique for constructing scalable and reliable multicast protocols. This evaluation has been done by simulating several gossip and non gossip-based multicast protocols in a network simulator NS2. The evaluation comprises of three parts: (1) a comparison of the gossip technique with other traditional message loss detection and recovery techniques, (2) an extensive performance evaluation of the gossip technique under several different operating parameters, and (3) an evaluation of how the performance of a gossip-based multicast protocol is affected by varying various gossip parameters. The main conclusion of the paper is that the gossip technique is certainly a useful technique for constructing scalable and reliable multicast protocols. However, it performs poorly under some specific operating conditions, and its performance is affected by the selection criteria used to select a gossip subgroup and gossip period. Based on the results of this experimental evaluation, the paper identifies three major factors that critically affect the scalability and reliability of a multicast protocol. These are (1) percentage of overhead messages exchanged, (2) overhead message concentration, and (3) application characteristics that determine message arrival rate and message arrival pattern.