S. El-Ansary, E. Aurell, and S. Haridi (Sweden)

PeertoPeer overlays, DHT performance, Structured Over lay networks, Data Collapse, Complex Systems

In the majority of structured peer-to-peer overlay networks a graph with a desirable topology is constructed. In most cases, the graph is maintained by a periodic activity per formed by each node in the graph to preserve the desir able structure in face of the continuous change of the set of nodes. The interaction of the autonomous periodic activi ties of the nodes renders the performance analysis of such systems complex and simulation of scales of interest can be prohibitive. Physicists, however, are accustomed to dealing with scale by characterizing a system using intensive vari ables, i.e. variables that are size independent. The approach has proved its usefulness when applied to satisfiability the ory. This work is the first attempt to apply it in the area of distributed systems. The contribution of this paper is two fold. First, we describe a methodology to be used for an alyzing the performance of large scale distributed systems. Second, we show how we applied the methodology to find an intensive variable that describe the characteristic behav ior of the Chord overlay network, namely, the ratio of the magnitude of perturbation of the network (joins/failures) to the magnitude of periodic stabilization of the network.

Important Links:

Go Back