Understanding Long-Term Evolution and Lifespan in Peer-to-Peer Systems

Abstract

Although peer-to-peer system is scalable for content distribution, its lifespan is shorter than traditional systems. More and more researchers began to pay close attention to lifespan and proposed specialized methods to extend it. However, what factors influence lifespan and how the effects are? To answer these questions, we analyze the long-term characteristics of swarms by our sampling measurement. The results show that swarms have different behaviors comparing with previous studies that focused on flash-crowd. Based on these new findings, we propose a long term model to depict the swarm evolution in large time scale. According to our model, the attenuation parameter of arrive rate and the average task length influence lifespan linearly while the initial arrival rate and peer availability have logarithmic influence. In order to validate our model, we compare them with real swarm and simulations. Our experiments show that the model captures the evolution closely and our results are valid in different situations.