【作者】 余一娇；

【导师】 金海；

【作者基本信息】 华中科技大学 ， 计算机系统结构， 2009， 博士

【摘要】 过于严重的搭便车行为会降低对等网络的健壮性、服务响应速度、可用性、生命周期等，如何抑制自私节点的搭便车行为是对等网络研究的重要问题。然而，现有对等网络搭便车抑制机制存在公平性欠佳、计算复杂性较高、工程实现难度大等问题。因此，对等网络搭便车抑制技术研究需设计新型搭便车抑制机制，模型化分析新型搭便车抑制机制的效率以及对网络服务性能的影响，并在真实对等网络系统中测试新型抑制机制的可用性和效率。基于文件复制的搭便车抑制机制通过复制过载节点中的高频访问文件到邻接搭便车节点，迫使搭便车节点为对等网络作贡献，提高对等网络的健壮性、文件服务的可用性等。提出复杂对等网络图定量描述对等网络的拓扑结构、节点状态和共享文件分布特征，分析搭便车抑制机制对搭便车节点比率、对等网络健壮性、文件共享服务可用性等的影响规律。通过在基于语义的对等网络科技文献共享系统SemreX中的应用，测试了基于文件复制的搭便车抑制机制的实用性和效率。针对各节点都期望享受高质量服务，又维持对等网络稳定运营的需求，提出了一种通过共享文件复制和负载均衡的搭便车抑制机制FRWBBA(File Replication andWorkload Balancing based Approach)。FRWBBA的思路是：各节点周期性监视自身资源利用率，判断节点是否过载；若节点过载，则主动向邻接搭便车节点透明地复制少数高频访问文件；更新共享文件信息，文件接收节点开始为对等网络提供更多服务。FRWBBA在不降低搭便车者所享受的服务质量前提下，迫使一些搭便车节点参与对等网络服务，保护了过载节点。过载节点如何实现高效的文件复制操作，是实现搭便车抑制机制FRWBBA的关键。根据对等网络文件访问次数分布近似服从齐普夫定律的特征，提出将过载节点中被访问次数最高的少量文件复制到邻接搭便车节点，该方案以较低对等网络通信开销、快速降低了过载节点的工作负载。提出将高频访问文件复制到邻接搭便车节点中负载最轻的部分节点，或根据文件访问路径规律，将热点文件复制到高频服务请求转发节点。仿真表明，最低负载优先策略和基于访问路径的目标节点选择策略，均是高效的文件复制目标节点选择策略。前者能快速消除对等网络中的过载节点，后者能显著降低文件查询服务的平均访问跳数。为了客观比较不同搭便车抑制机制的效率差异，归纳出搭便车抑制机制的效率评价核心指标集；提出复杂对等网络图来定量描述搭便车抑制机制对对等网络状态的影响。复杂对等网络图描述了对等网络的拓扑结构、节点异质性和共享文件分布状态等特征，定义了搭便车抑制机制的效率评价核心指标集。应用复杂对等网络图和无标度网络理论，分析了FRWBBA对搭便车节点比率、服务可用性、对等网络健壮性等的影响规律；比较了FRWBBA和基于激励的搭便车抑制机制的效率差异，得出FRWBBA是更受节点用户和对等网络运营者欢迎的搭便车抑制机制。为了测试FRWBBA的工程可行性和效率，用Java实现了FRWBBA，并封装为AntifreeridingWare软件包。基于语义的对等网络科技文献共享系统SemreX采用AntifreeridingWare抑制对等网络中的搭便车行为。该应用实例表明AntifreeridingWare有效、可行，且易于在第三方对等网络文件共享系统中应用。与基于激励的对等网络搭便车抑制机制相比，FRWBBA维护了对等网络服务的公平性、保证了对等网络尽力为全体用户提供服务；与基于博弈论的搭便车抑制机制相比，FRWBBA无需对等网络节点行为的全局信息，易于实现且可扩展性强；与采用社会网络和经济模型的搭便车抑制机制相比，FRWBBA更具工程可行性。更多还原

【Abstract】 Most of peers in Peer-to-Peer (P2P) networks merely take what they need from others,and only a few altruistic peers provide services. These selfish behaviors are regarded asfree riding, and selfish peers are referred to as free riders. Free riding is a great challengefor P2P networks because it impairs robustness, availability, and lifetime of P2P systems.Some anti-free riding mechanisms are proposed, among which, the incentive mechanismsnot only degrade the Quality of Service (QoS) of free riders, but also decrease the numberof online users; the game theoretical approaches are too complicated to implement inlarge-scale P2P networks; the employment of social network/economic models in practiceis not fully investigated yet. Thus, a feasible, scalable and simple anti-free ridingmechanism is required.This thesis is dedicated to a distributed anti-free riding mechanism. The salient featureof our research is that peer’s heterogeneity and inherent selfishness is given greatconsideration. P2P network is described with a novel model, which includes the networktopology, peers status, and files distribution. The efficiency criteria are defined formallyand clearly with the novel P2P network model, and different anti-free riding mechanismsare compared. Finally, the novel anti-free riding mechanism is employed in a scientificreferences sharing system, which shows that it is feasible and efficient.A file replication and workload balancing based anti-free riding mechanism (FRWBBA)is proposed in Chapter 2. The basic idea of FRWBBA is as follows. Every peer monitorsits workload status periodically and independently. If the workload is too heavy, theoverloaded peer transfers some frequently accessed files to neighbored free ridersproactively and transparently. As a result, these neighbored free riders share more popularfiles and make more contributions to P2P systems. The number of free rider is reduced,while the overloaded peers are well protected. Extensive simulations are carried out toverify the efficiency of FRWBBA. The results show that FRWBBA makes the whole P2Pnetwork robust and the shared files distributed appropriately.The strategies of file replication from overloaded peer to neighbored free riders areinvestigated in Chapter 3. As Zipf’s law is suitable for modeling P2P file-accessedfrequencies, replicating the most popular files from overloaded peer to neighbored freeriders can decrease the workload effectively. Three strategies of target peers selection in terms of workload are proposed. Simulation results indicate that popular files should bereplicated to the least workload neighbors, which can eliminate overloaded peers rapidly.As an alternative, overloaded peer can transfer popular files to the ones near the serviceconsumers to reduce the hops of file access.P2P network modeling is important to evaluate anti-free riding mechanisms, which isdiscussed in Chapter 4. The essential metrics for P2P networks with respect to free ridingare identified. On this basis, a complex P2P graph is proposed, which can describe the P2Pnetwork topology, the heterogeneity of peers, and the relationship between shared filesdistributed in peers and the network topology. Since P2P networks are scale-free, theirrobustness and crispness is analyzed. The efficiency of FRWBBA and incentivemechanisms is evaluated by using the complex P2P graph and properties of scale-freenetworks. FRWBBA is shown superior to incentive mechanisms from the perspectives ofboth P2P users and suppliers.FRWBBA is implemented in JAVA and packaged as AntifreeridingWare. The softwaredevelopment methods of AntifreeridngWare are presented in Chapter 5. To illustrate theusage of the package, it is integrated into Semrex, a P2P scientific references sharingsystem, to deal with free riding. The development experiences suggest that anti-free ridingmethod cannot be separated from P2P applications completely, because it needs thesupports from P2P programs.The model analysis, Simulations, and applications in real P2P system all validate thatFRWBBA is an efficient anti-free riding approach. FRWBBA reduces the number of freeriders, and makes P2P networks provide services to all peers in best efforts. The abundantfiles in P2P network are maintained and distributed properly after anti-free riding control.Compared to the incentive mechanisms, FRWBBA offers better fairness to those peersbehind firewall and Network Address Translations, or peers with low speed Internetconnections. The software development experiences of FRWBBA help it be integrated intodifferent P2P applications quickly.更多还原