Performance Analysis And Design For File Transfers In Peer-to-Peer File Sharing Networks

Peer-to-peer file sharing applications, which are the major applications of peer-to-peer networks, are popular in Internet in recent years. Peer-to-peer file sharing networks attract many researches, since they can fast disseminate large files to many users by providing cooperative file searching and distributed file sharing. This dissertation focuses on the performance of file transfer in peer-to-peer file sharing networks, which is one of the major issues of the research and design of peer-to-peer networks. Our research includes the algorithm designed for fast file dissemination, the systematic analysis on the performance of file transfer, and a system design for fast file dissemination in mobile environments. Our major works and contributions are summarized as follows.(1) The research on the problem of fast disseminating a large file from a single source to a large number of users. How to decrease the dissemination time by fully utilizing the upstream bandwidth of nodes is analyzed, while an algorithm called Receiver-Schedule is presented for fast file dissemination.How to minimize the time of disseminating a large file from a single source to a large number of users is a fundamental question when examining the performance of file transfer in peer-to-peer applications. Previous works fail to show that how close the dissemination time of their algorithms are to the optimal solution. In this paper, it is analyzed that how to decrease the dissemination time by efficiently utilizing the upstream bandwidth of nodes. A tight lower bound on the dissemination time is presented, which can serve as a benchmark for the performance of peer-to-peer algorithms, while the bound is achievable in two cases. Based on the assumption that only the upstream bandwidth of nodes can constrain the transfer rates between nodes, an equivalent condition along with a sufficient condition for fully utilizing the upstream bandwidth of nodes are presented. Based on the theoretical analysis, an algorithm called Receiver-Schedule is designed, which can dynamically adjust the chunk distribution among nodes, for fast dissemination. Simulations show that the dissemination time is more than the lower bound by at most four percent in many cases when using the algorithm.Our works well solve the problem of minimizing the dissemination time to a large extend.(2) The analysis on the impact of nodes'behavior on system throughputs in peer-to-peer file sharing networks.It is essential to analyze the impact of nodes'behavior on system throughputs, when examining the performance of file transfer in peer-to-peer file sharing networks. The system throughput can be measured by the sum of the file data that is downloaded by all the nodes in a network per unit time. We examine the impact of nodes'behavior, including their online time and greed of downloading files, on system throughputs by modeling and simulations. We extend an existing peer-to-peer network model and classify nodes according to their behavior. Two major kinds of peer-to-peer architectures are focused on: centralized peer-to-peer networks and structured peer-to-peer networks. It is found that when the non-free-riders in a network are all greedy in downloading files, the system throughput has little room to increase while the non-free-riders'throughput degrades badly with the increasing percent of greedy free-riders in the network. Here a free-rider is a node that downloads files from its peers but does not share files to other nodes. When all the non-free-riders are non-greedy with long average online time, however, the system throughput has much room to increase and the non-free-riders'throughput degrades little with a high percent of greedy free-riders in the network. It is also found that if a peer-to-peer network can tolerate a high percent of greedy free-riders without suffering much throughput degradation, the network must contain some non-greedy non-free-riders that contribute great idle service capacity to the network.Our work reveals the typical impacts of nodes'behavior on system throughputs, and explains the cause and preconditions for the impacts. It helps people to further understand the performance of file transfer in peer-to-peer networks.(3) The analysis on the service rate for transferring a file in peer-to-peer file sharing networks.When examining the file transfer performance in a peer-to-peer file sharing network, a fundamental problem is how to describe the service rate for transferring a file. In this paper, the problem is examined by analyzing the distribution of nodes'upstream bandwidth among their concurrent uploading transfers. On the service rate that a receiver obtains for transferring a file, a sufficient condition for it to asymptotically be uniform is presented. On the aggregate service rate for transferring a file in a network, a sufficient condition for it to asymptotically follow a Zipf distribution is presented. These asymptotic equalities are both in the mean square sense.These analysis and the sufficient conditions provide a mathematic base for modeling file transfer processes in peer-to-peer file sharing networks.(4) A design of peer-to-peer file sharing system for UMTS networks, which achieves fast file dissemination in the mobile environment.In UMTS (universal mobile telecommunications system) networks upgraded with HSPA (high speed packet access) technology, the high access bandwidth and advanced mobile devices make it applicable to share large files among mobile users by peer-to-peer applications. To receive files quickly is essential for mobile users in file sharing applications, mainly because they are subject to unstable signal strength and battery failures. While many researches present peer-to-peer file sharing architectures in mobile environments, few works focus on decreasing the time spent in disseminating files among users. We present an efficient peer-to-peer file sharing design for HSPA networks called AFAM - Adaptive efficient File shAring for uMts networks. AFAM can decrease the dissemination time by efficiently utilizing the upload-bandwidth of mobile nodes. It uses an adaptive rearrangement of a node's concurrent uploads, which causes the count of the node's concurrent uploads to lower while ensuring that the node's upload-bandwidth can be efficiently utilized. AFAM also uses URF - Upload Rarest First policy for the block selection and receiver selection, which achieves real rarest-first for the spread of blocks and effectively avoids the"last-block"problem in file sharing applications. Simulations show that, AFAM works well in mobile environments and achieves much less dissemination time than other existing protocols.