CAREER: Pricing and Resource Allocation in Multiservice Broadband Communication Networks

新的技术进步和研究方面的最新发展正在引导人们走向增强的下一代互联网。这种新的媒体将超越目前的"尽力而为”能力，并发展成为一个多服务网络，能够适应不同的服务类别，以支持各种类型的应用程序和业务需求。现在，技术已经发生了巨大的变化，计算机网络已经变得无处不在，拟议的工作旨在解决一些重要的挑战：（i）有效地使用网络资源，（ii）促进创造一个健康的市场环境，新的网络服务可以引入和维持。 为此，建议（1）设计机制和工具，使网络服务提供商能够优化其运营并使其在经济上可行，以及（2）通过定价向网络服务的用户提供足够的激励，这将有助于平滑需求高峰并导致更有效地利用可用资源。 互联网和一般的计算机网络正在发展成为一个真实的产业。其他较成熟行业的发展（例如，航空公司、汽车租赁）都预示着它的未来。这些行业实施复杂的收入管理做法，包括定价和有效的资源分配。在网络中，这些行动也有助于缓解拥塞，拥塞不仅是由于缺乏带宽造成的，而且越来越多地是由于Web服务器过载造成的。 拟议的研究包括以下方面的具体计划：（1）将用户对价格的反应纳入多种网络服务的定价模型（需求函数）、服务特征、网络影响以及问题中固有的不确定性;（2）根据测量估计各种服务消耗的网络资源量，这可能有助于定价和其他资源分配机制;（3）优化服务器的运行Web，在服务器集群中实现复杂的任务调度和路由;以及（4）做出战略性服务供应决策，这些决策受竞争和消费者选择的影响，包括容量规划、在带宽市场中购买或销售，以及决定所提供服务的最佳菜单。 拟议的工作将利用主要研究者具有相当专业知识的方法和技术，在某些情况下，这些方法和技术已帮助开发。这些方法包括：（i）大偏差，（ii）动态（和近似动态）规划，（iii）势函数方法。 在教育方面，拟议的计划旨在促进优化和随机技术之间的协同作用，并准备研究生精通这两个领域谁可以为这一研究议程作出贡献。它包括具体的（研究生和本科生）课程开发思路，邀请研讨会系列的组织，并建立一个教学软件实验室，将提高本科生的招聘和教学。PI组的研究生通常在随机过程，控制和优化方面有很强的背景。两个拟议的计算机网络研究生课程旨在帮助他们集中在这一领域的研究，提供了足够深入的调查进展和悬而未决的问题。

New technological advances and recent developments in research are leading the way towards an enhanced next-generation Internet. This new medium will surpass the current best effort only" capability and evolve into a multiservice network able to accommodate differentiated classes of service to support various types of applications and business requirements. Now that the technology has dramatically evolved and computer networks have become ubiquitous, the proposed work aims at addressing some of the important remainingchallenges: (i) to use the network resources efficiently, and (ii) to facilitate the creation of a healthy market environment, where new network services can be introduced and sustained. To this end, it is proposed (1) to devise mechanisms and tools that will enable network service providers to optimize their operations and make them economically viable, and (2) to provide, through pricing, sufficient incentives to users of network services that will help smooth spikes in demand and lead to a more efficient utilization of the available resources. The Internet, and computer networking in general, are evolving into a real industry. The developments in other more mature industries (e.g., airlines, car rentals) are indicative of its future. Such industries implement sophisticated revenue management practices including pricing and efficient resource allocation. In networking, these actions can also be instrumental in alleviating congestion which is not only caused by the lack of bandwidth but, increasingly, due to overloading of Web servers. The proposed research contains specific plans on (1) models for pricing multiple network services that incorporate users' reaction to prices (demand functions), service characteristics, network effects, and the inherent uncertainty in the problem; (2) estimating from measurements the amount of network resources consumed by various services, which can be instrumental in pricing and other resource allocation mechanisms; (3) optimizing the operation Web of servers, i.e., implementing sophisticated scheduling and routing of tasks in clusters of servers; and on (4) making strategic service provisioning decisions, whichare affected by competition and consumer choice and include capacity planning, buying or selling in bandwidth markets, and deciding the optimal menu of offered services. The proposed work will draw upon methodologies and techniques that the PI has considerable expertise and, in some instances, has helped in developing. These include: (i) large deviations, (ii) dynamic (and approximate-dynamic) programming, and (iii) potential function methods. On the educational front, the proposed plan aims at promoting a synergy between optimization and stochastic techniques, and at preparing graduate students well versed in both areas who can contribute to this research agenda. It consists of specific (graduate and undergraduate) curriculum development ideas, the organization of an invited seminar series, and the creation of an instructional software laboratory that will enhance undergraduate recruitment and teaching. Graduate students in the PI's group typically have strong background in stochastic processes, control, and optimization. Two proposed graduate courses in computer networking aim at helping them focusing their research in this area by providing a sufficiently deep survey of advances and outstanding issues.