Research On Distributed Cache Strategies In Peer-to-Peer Networks

Peer-to-Peer (P2P) is a distributed network which is an overlay on the physical network. In P2P, each computer has the same function, and they share resources from each other. Although there are a large number of studies, the problem of poor efficiency of P2P still exists. Moreover, with the rapid expansion of Internet and users, P2P networks have become the largest bandwidth consumers on the Internet. In order to improve search efficiency and reduce traffic of Internet backbone, the caching is an effective solution. Despite the centralized caching based on the central servers has been mature, but the distributed caching mechanism for P2P systems is still in the initial stage.The search mechanism used in unstructured P2P networks is a blind random probe. This strategy has lower search efficiency. In this paper, we design an efficient network topology and search algorithm, and propose novel and effective caching placement and replacement strategies for P2P caching. We use distance-based network topology to cluster the topologically close peers, and take preference-based three-way random walk to increase local query and achieve better searching performance. We also utilize remote peer caching placement approach to determine the locations of the objects needing to be cached in order to improve caching effectiveness. Our high popularity and least request caching replacement algorithm effectively relieves the over-caching problems for popular files and offers satisfactory caching performance for other files. We compare our design with various common and heuristic caching algorithms and the results show that combined with proposed underlay network infrastructure, our caching strategies can deliver better query hit rates, smaller query delay, higher cache hit rates, and lower communication overhead.In order to improve the overall performance of structured P2P networks, we propose a distributed caching approach bashed on structured P2P networks. We use a hierarchical DHT (Distributed Hash Table) infrastructure to select an adequate number of cached copies for the object with different popularity and determine the locations of these copies carefully in order to improve cache effectiveness. We compare our design with various common and heuristic caching algorithms by conducting extensive simulation experiments. We observe that combined with the hierarchical query infrastructure, our caching strategies can deliver lower query delay, better load balance and higher cache hit ratios. Our algorithms effectively relieve the over-caching problems for the most popular objects and offer satisfactory caching performance for other types of objects. Most of the P2P distributed caching mechanism is a passive cache, that is, only when a file is requested will trigger cache operations, which will result in less cached copies for rare objects because the request frequency of rare objects is relatively low. However, in many cases, rare objects are very valuable, such as the rare blocks in P2P file sharing systems and P2P streaming applications. In order to improve the search efficiency of rare objects, we propose proactive caching strategy and rare object search algorithm which needs less number of cached copies. Thus, we can reduce the overhead caused by the caching mechanism. The simulation results show that the proactive caching algorithm has better efficiency for rare objects, search hit rate reaches at 98%; and our strategies can maintain lower communication overhead, which can reduce nearly 40% traffic compared to other algorithms.P2P-based streaming media applications and P2P file sharing systems have become increasingly popular. These applications are becoming ever larger, and more important impact on the Internet. However, we are facing with several challenges that stem from the commercial nature of these systems. The protocol, architecture and algorithm are proprietary, which makes it very hard to analyze the system and figure out the problems and bottlenecks. Most research studies used the black box approach to measure and evaluate system properties such as upload/download performance, user distribution, and session lengths, etc. In order to cover more measurements, we propose a universal measurement model for P2P networks after researching on other network measurement mechanism. And we use the proposed model to measure P2P streaming channels with different popularity to validate our design. The measurement metrics include data transfer rate, cache mechanism, peer geographic distribution, session length. These measurement results not only can provide support for distributed caching, but also for other researchers.