© The Author 2013. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved. For Permissions, please email: journals.permissions@oup.com Advance Access publication on 6 March 2013 doi:10.1093/comjnl/bxt022 A Node-Link-Based P2P Cache Deployment Algorithm in ISP Networks Haibin Zhai 1,3,∗∗ , Albert K. Wong 2 , Hai Jiang 1 , Yi Sun 1 , Jun Li 1 , Zhongcheng Li 1 1 Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China 2 Hong Kong University of Science andTechnology, Hong Kong, China 3 Graduate University of Chinese Academy of Sciences, Beijing, China ∗ Corresponding author: zhaihaibin@ict.ac.cn Peer-to-peer (P2P) systems are imposing a heavy burden on internet services providers (ISPs). P2P caching is an effective way of easing this burden. We focus on the cache deployment problem as it has a significant impact on the effectiveness of caching. An ISP backbone network is usually abstracted to a graph comprising nodes representing core routers and links connecting adjacent core routers. While deploying P2P caches at nodes (NCD, node-based cache deployment) can reduce the amount of P2P traffic transmitted from access networks to the ISP backbone network, deploying P2P caches on links (LCD, link-based cache deployment) can directly reduce the amount of P2P traffic on the ISP backbone network. However, neither NCD nor LCD maximizes the performance of P2P caches. In this paper, we propose a node-link-based cache deployment method (NLCD), which optimally selects nodes or links as deployment locations during the cache deployment process. First, we propose an analysis model and define an optimal cache deployment problem for NLCD. Then, we prove that this problem is NP complete and develop a corresponding deployment algorithm. Experimental results show that the average link utilization of NLCD is 5–15% lower than that of LCD, and 7–30% lower than that of NCD. Keywords: peer-to-peer network; peer-to-peer traffic cache; cache deployment algorithm Received 30 July 2012; revised 7 February 2013 Handling editor: Jongsung Kim 1. INTRODUCTION Peer-to-peer file-sharing systems such as BitTorrent, eDonkey and eMule have become popular in recent years. Reports by CacheLogic [1] and Ipoque [2] have indicated that P2P file-sharing traffic (called P2P traffic hereafter in this paper) accounted for 60–75% of the worldwide Internet traffic in 2005 and 43–70% in 2009, respectively. A report by China Telecom [3] has indicated that P2P traffic accounted for about 55% of the China traffic in 2010. P2P traffic is consuming a large amount of network resources and imposing a great burden on internet services providers (ISPs) worldwide. Several approaches have been proposed to ease the burden imposed by P2P traffic. These include traffic limiting [4], traffic locality [5–7] and traffic caching [8–16]. Compared with the first two approaches, caching is completely transparent to the clients and does not require changes in the P2P software. The benefits of caching for P2P traffic have been studied in many papers. Karagiannis et al. in [14] identify that current P2P solutions are ISP unfriendly. They suggest that deploying caches to reduce the load on ISP networks. Gummadi et al. in [15] investigate the differences between web caching and P2P caching. They believe that caching has a stronger opportunity to benefit P2P systems than it has in the web. Hefeeda et al. in [9] propose two models of cooperative caching for P2P traffic and analyze the potential gain of cooperative caching. Besides the benefits of P2P caching, caching algorithms [8, 10, 16] and measurement studies of P2P systems [10, 15] have also been reported. Caching of P2P traffic may also raise legal issues. However, discussing these legal issues is beyond the scope of this paper. Although the deployment strategy for P2P traffic caches can have a significant impact on their effectiveness, the problem has started to receive attention only in recent years [11–13]. Ye et al. in [11] study the link-based cache deployment (LCD) strategy, and Kamiyama et al. in [12, 13] study the node-based The Computer Journal, Vol. 57 No. 2, 2014 at Pennsylvania State University on March 6, 2016 http://comjnl.oxfordjournals.org/ Downloaded from