Server Selection in Emerging Information Delivery Environments

从内容服务器到客户端的信息及时交付现在显然是无处不在的网络的主要角色之一。 随着通过网络寻求信息的人数的增加，信息传递服务的可扩展性成为一个重要的问题。 服务器复制是服务提供商用来满足不断增长的信息需求的一种常用技术。 然而，有效的服务器复制需要适当的服务器选择方法，否则将无法实现复制的全部好处。 糟糕的服务器选择通常会导致一系列相关的系统性能问题：客户端感知到的性能下降，由于服务器过载而可能拒绝请求，访问所选服务器的其他客户端观察到的性能下降，以及其他网络连接由于不平衡负载导致的热点而观察到的较差性能。本提案的重点是服务器选择的设计和评估方法，特别是在广域服务器复制的上下文中。 我们提出了一个研究计划，提供了一个系统的调查服务器选择问题的服务和网络的当前状态开始，但主要关注未来的网络和服务环境。 广域网能力和信息服务性质的不断发展和创新将导致服务器选择问题更加复杂，影响选择的性能和政策问题更加多样化。 这项工作将考虑五种类型的服务：传统的Web和FTP服务、多播服务、多媒体流服务、多源服务和中介服务。 研究人员将这些视为不同的服务维度，这些维度可能以某种组合存在于部署的服务中，并探索了每个维度的服务器选择挑战。 研究人员还将考虑如何使用新的网络基础设施功能（从支持某种形式的差异化服务到更深远的主动网络建议）来增强和支持服务器选择功能。开发和评估使用网络支持的技术，以帮助服务器选择功能。组播服务器选择问题的特征、机制、策略及其评价.在当前和未来网络环境中选择流视频服务器的技术.使用交互式多媒体服务器（IMJ）（imj.gatech.edu）和研究人员以前开发的多播Web服务器，为多播和多媒体流服务器的建议服务器选择协议提供概念证明。 5.进一步表征和服务器选择建议两种新的服务模式（多源和代理服务）。一个可重复使用的实验平台，用于试验服务器选择的建议。建议的工作将超越服务器选择机制，在今天的环境中，不需要是复杂的，轮循或其他简单的技术，用于分配客户端到服务器是目前足够的。然而，在上述趋势将导致的更丰富的环境中，这种简单和特别的技术将无法很好地工作。 这项工作考虑的体系结构，机制和政策，服务器的选择可能会在这个未来的网络服务环境中完成。 研究人员认为，他们的研究实际上将发挥促进作用，因为如果没有一个可行的服务器选择系统，就无法实现这一未来的好处。

Timely delivery of information from content servers to clients is now clearly one of the primary roles of ubiquitous networking. As the number of people seeking information through networks increases, scalability of information delivery services becomes an important issue. Server replication is one common technique used by service providers to accommodate the increasing information demands. Effective server replication, however, demands appropriate server selection approaches, without which the full benefits of replication will not be realized. Poor server selection often leads to an array of associated system performance implications: degraded performance as perceived by the client, possible rejection of the request due to server overload, degraded performance observed by other clients accessing the selected server, and poor performance observed by other network connections due to hot spots resulting from unbalanced load.The focus of this proposal is on the design and evaluation of server selection approaches specifically in the context of wide-area server replication. We propose a research program that provides a systematic investigation of the server selection problem starting with the current state of services and networks but mostly concerned with future networking and service environments. Continuous evolution and innovation in the capabilities of wide-area networking and the nature of information services will result in a much more complex server selection problem with a wider diversity of performance and policy concerns that affect selection. The work will consider five types of service: traditional Web and FTP services, multicast services, multimedia streamed services, multi-source services and brokered services. The researchers consider these as different service dimensions that may be present in some combination in deployed services and explore the challenges in server selection for each dimension. The researchers will also consider how new network infrastructure capabilities (ranging from support for some form of differentiated service to the more far-reaching Active Networking proposals) can be used to enhance and support the server selection function.The main contributions of the research will be:1. Development and evaluation of techniques for the use of network support to aid the server selection function.2. Characterization, mechanisms, policies and their evaluation for the multicast server selection problem.3. Techniques for streamed video server selection in current and future network environments.4. The use of the Interactive Multimedia Jukebox (IMJ) (http://imj.gatech.edu) and the researchers' previously developed multicast web server to provide proof of concept for the proposed server selection protocols for multicast and multimedia stream servers. 5. Further characterization and server selection proposals for two novel service paradigms (multiple-source and brokered services).6. A reusable experimental platform for experimenting with server selection proposals.The proposed work will go beyond server selection mechanisms in today's environment which do not need to be complex; round-robin or other simple techniques for allocating clients to servers are currently adequate. Such simple and ad-hoc techniques will not, however, work well in the richer environment that the trends above will lead to. The work considers the architecture, mechanisms and policies by which server selection may be accomplished in this future networked service environment. The researchers believe that their research will, in fact, play an enabling role, since this future's benefits cannot be realized without a workable system for server selection.