Görevdeş ağlarda topolojinin bilgi yayılımı ve ara bellek yönetimine etkisi

Keeping every node updated about the newly generated data in the dynamic and rapidly changing environment of a network is achieved via different data dissemination algorithms. Epidemics is one of the widely accepted algorithms because of its reliability and robustness. Message loss recovery for maintaining the reliability of the content delivery in case of message losses is achieved via several buffer management techniques. Efficient usage of limited memory resources is the basic deal for buffer management.In this thesis, we present our analysis of peer-to-peer (P2P) networking phenomena, namely data dissemination and buffer management, focusing on topological perspectives. For data dissemination, we examine spreading of epidemics for anti-entropy algorithms on several overlay network topologies, considering peer proximity. We derive nodes' exact probability distributions of being infected in each epidemic cycle of data dissemination. For buffer management, we examine buffering with an efficient algorithm, Stepwise Fair-share Buffering, that uses memory resources effectively and distributes the buffering load uniformly throughout the system. We analyze the effect of different topologies on buffer management, using hierarchical and power-law topologies, two basic types of topology modeling the Internet.For data dissemination, the effect of topological properties is studied using numerical evaluations. The rate of dissemination is found to be related to the adjacency matrix in a nonlinear way. For buffering, performance evaluations of various models with hierarchical and power-law topologies are conducted. Scalability, reliability, dissemination delays and uniformity are considered as basic performance parameters. We have shown that Stepwise Fair-share Buffering method facilitate better uniformity in distribution of buffering load, in view of our simulations. We expect to have higher delays due to decision process performed for bufferer selection; however, it is also shown that dissemination delay performance drawback is eliminated when power-law topologies are considered.