Computer Science Department

Abstract

One of the m ost im portant cha lleng es of peer- to-peer m ulticast pro tocols is the ability to efficiently deal with the high degree of tr ansiency inherent to their environm ent. As multicast functionality is pushed to au tonom ous, unpredictable peers, significant perform ance losses can result from group m embership changes and the higher failure rates of end-hosts when com pared to router s. Achieving high delivery ratios without sacrificing end-to-end latenc ies or incurring additional costs has proven to be a challenging task. This pape r introduces Nem o, a novel pe er-to-peer m ulticast proto col that aim s at achieving this elusive goal. Based on two sim ple techniques: (1) co-leaders to minimize dependencies and, (2) triggered negative acknowledgments (NACKs ) to detect lost packets, Nemo’s design em phasizes conceptu al sim plicity and m inimum dependencies, thus ach ieving perform ance characteristics capab le of withsta nding the na tural instability of its target environm ent. We present an extensive com parative eval uation of our protocol through sim ulation and wide-area experim entation. W e compare the scalability and perform ance of Ne mo with that of three alternat ive protocols: Narada, Nice and Ni ce-PRM. Our results show that Nemo can achieve delivery ratios (up to 99.9%) similar to those of com parable protocols under high failure rates, but at a fraction of their cost in term s of duplicate packets (reductions > 90%) and control-related traffic (reductions > 20%).