【作者】 李慧霸； 彭宇行； 卢锡城；

【Author】 Huiba Li1+, Yuxing Peng1, Xicheng Lu1 1(Department of Computer Science, National University of Defense Technology, Changsha, Hunan province, 410073, China)

【机构】 国防科技大学计算机学院；

【摘要】 在P2P系统中,源节点的选择方法对网络的性能有很大影响,已有的拓扑相关源节点选择法着眼于改进在请求节点执行的算法,忽视了服务端的源节点选择质量,可能会导致请求节点根本无法选出满意的服务节点。本文提出了适用于执行查询请求的服务端的源节点选择算法,可以使节点间平均延迟大约降低80%,同时也具有可接受的执行速度。本算法最大的特点是简单、有效、高效。更多还原

【Abstract】 In P2P networks, the algorithm of selecting source peers has a significant influence on the performance of the system. Existing methods on topology-aware peer selection focus on improving the algorithms performed in the requesting peers, these methods ignore the quality of source peer selection in server side, and this might cause the requesting peers to be unable to pick out satisfied servant peers. We propose a source peer selection algorithm suitable for the server side which executes query requests. While keeping an acceptable performance, this approach is able to reduce the average delay between peers by about 80%. The key feature of this approach is simple, efficient and effective. First of all, we utilize a synthetic n-dimensional Euclidean coordinate for each peer to estimate the network distance. Secondly, the n-dimensional Euclidean space is splite into a grid, and only peers in nearby region are considered in the following selection. Finally, here comes the problem of selecting the best m source peers out of n candidates. To solve this problem, we use the partition operation from the quick sort algorithm to constitute a quick selection algorithm, which have an averge time complexity of O(n). Our experiment consists of 2 parts, the speed test and the effect testification. In the speed test, we concentrate on the core selection algorithm, and test the average time consuming of the algorithm, where the number of candidates ranges from 128 to 1M. The speed of selecting sources from 100,000 candidates on a ordinary PC is 1ms on average. In the effect testification, we generate a number of distinct network topology structure by using GTITM together with BRITE, and we assume that network packets are deliveried between nodes only alone the path with lowest delay. The result shows that our approach is able to reduce the average delay between peers by about 80%, compared to random selection.更多还原