【作者】 张冠湘；

【导师】 杨宗凯；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2005， 博士

【摘要】 近年来因特网的发展呈现出网络资源相对稀缺、应用多样化和商业化几大特点。在商业化环境下仅仅通过超量供给来解决网络资源稀缺和满足不同类型用户的服务要求不可行。所以如何合理分配相对稀缺的网络资源、激励用户选择合适的服务类型, 从而优化网络整体的效益, 便成为一个重要的研究课题。传统的网络资源分配方法大多数从工程方面来研究怎么来提高网络效率,而没有从用户的角度来分析,所以不能达到资源的充分利用。近年来,国外的研究者开始尝试着把微观经济学研究方法引入网络资源分配中,为这一领域的研究提供了崭新的思路。这种新方法在网络资源分配中引入用户效用的概念,利用供求相互作用机制、价格调节机制等微观经济学理论来求得网络资源的最优分配。这一方法在网络资源分配中可以起到在以下四个方面的作用:(1)建立相对统一的网络资源分配评价指标; (2)计费与网络速率控制机制相结合作为一种优化网络资源分配的手段; (3)计费在多服务提供中激励用户行为; (4)计费是网络商业化运行的基础。当前大部分网络带宽优化分配的工作基本只适于单播环境,组播网络的异构性问题使得组播网络中的资源优化分配问题要比单播环境复杂得多,组播的优化带宽尚未得到很好的解决; 而当前互联网将由提供单一的尽力而为服务向提供区分服务的方向发展,多服务网络计费在建模方面和机制设计方面尚有许多工作需要展开。本文将计费机制应用于网络资源分配中,围绕组播网络中的带宽优化分配机制和优先级网络中的优先级分配和计费机制展开了研究。本文的工作得到国家自然科学基金“基于Internet 的交互式流媒体分发技术研究”(No.60302004)的资助。本文取得的研究成果包括如下几个方面: (1) 提出了一种单速率组播速率控制与修剪算法。从经济而不是工程的观点来分析组播内部公平性,提出了一种基于效用的组播内部公平性标准。从速率控制和组播树修剪两方面来最大化组播会话的净收益,在分枝节点效用的基础上提出了一种基于动态规划的单速率组播速率控制与修剪算法,该算法具有可扩展性。更多还原

【Abstract】 In recent years, the development of Internet takes on three important features: the scarcity of resources, a wide variety of applications and commercialization. Overprovision can neither solve the problem of resource scarcity and nor meet differentiate service requirements,so how to optimally allocate the scarce network resources and incent users to select appropriate service types is an important issue. Traditional research on of network resource allocation focused on the improvement of engineering efficiency, but not took the satisfaction of users into consideration. As a result the network resource is not optimally utilized. The recent research on network resource allocation based on microeconomic theory brings us a new approach, where utility is introduced as an important concept. Optimal resource allocation is realized by the demand-supply mechanism and pricing mechanism. The new approach can be utilized in the following four aspects. First, utility based evaluation can be used as a comparatively unified evaluation criterion for network resource allocation. Secondly, combined with rate control and call admission control, pricing can be used as an optimal resource allocation mechanism. Thirdly, pricing can incent users to choose the appropriate service in a network with service differentiation. Finally pricing is the base of a commercial network. Up to now, however, most research on pricing-based rate control deals with the unicast scenario. The optimal rate control of multicast scenario is more complex because of the heterogeneity of multicast receivers. Today’s Internet only provides best effort service, and it is becoming more and more a multi-service network, the modeling and mechanism design of multi-service pricing are important issues to be dealt with. This thesis addresses on bandwidth allocation in multicast network and priority allocation in multi-service network, where pricing mechanism plays an important role. The work in this thesis has been supported by the National Science Foundation of China “Internet-based interactive streaming delivery technique”(No.60302004). The dissertation contributes in the following aspects. (1) A rate control and pruning algorithm for single-rate multicast sessions is proposed. Unlike the previous inter-receiver fairness criteria, a utility based fairness criterion is proposed. Based on the utility of junction node, a rate control and pruning algorithm is proposed to maximize the net benefit of the session. The algorithm is dynamic programming based and scalable. (2) A convex model is proposed for the bandwith allocation of network both with unicast and multirate muticast sessions, and an optimal multi-rate multicast rate control algorithm is proposed based on the utility of junction node. The optimal object of resource allocation is to maximize the social welfare. The algorithm is based on dual theory and gradient projection algorithm, where price is used to transfer congestion information to users. Compared with previous related work, the cost of links and discrete receiver rate of multicast are considered. An important step in the resource allocation iterative algorithm is to obtain the optimal rates which maximize the group net benefit. To address this problem a dynamic programming based algorithm is proposed based on the utility of junction node. The algorithm is distributed and scalable. (3) The incentive compatible problem of priority service is discussed. The optimal pricing strategy of preemptive priorities and non preemptive priorities is analyzed. It is proved that the optimal pricing strategy in implicit supply model is incentive compatible when the service is explicitly supplied. Compared with others’work, it indicates that the incentive compatible pricing is related with the user model. (4) Two auction-based priority allocation mechanisms are proposed. An infinite priority auction mechanism is proposed to solve the problem of priority allocation problem when users are with a continuous distribution of delay cost. The equilibrium bidding strategy and reserve price are discussed. An incentive compatible aggregate priority auction mechanism is proposed. Every user’s strategy-proof strategy is to announce the true value in the mechanism. The higher priorities are allocated to delay-sensitive users, and as aresult, the total delay cost of all users is minimized.更多还原