CAREER: Flexible Strategies for Internet Content Distribution

点击翻译按钮获取中文摘要

The emergence of new technologies for content distribution including content delivery networks (CDNs, end system multicast, distributed content exchange and application-level anycast are fundamentally changing the relationships between consumers and producers of digital content. Whereas traditional approaches to content delivery are predicated on point-to-point connections from a client to a single origin server, these new mechanisms for content distribution fundamentally undermine the assumptions inherent in that model. For example, CDNs redirect clients to a number of equivalent replica servers, while end system multicast goes further and blurs the distinction between client and server by requiring that every participant who wishes to access content serve content as well. To illustrate this point, consider the access graph plotting the flow of TCP connections serving a popular document through the network at an instant in time. Until very recently, this graph could easily be characterized as forming a hub-and-spoke topology with the hub at the origin server; now its complex structure is little known except perhaps to network engineers within the CDNs themselves. Broadly, the researcher's interests lie in characterizing the impact that new and increasingly distributed content delivery mechanisms will have on network traffic, resource management and the design of network protocols. In the research component of the researcher's career development program, he focuses on strategies for addressing the challenges and opportunities that distributed content delivery mechanisms will place on reliable transport protocols. In particular, the complex one-to-many and many-to-many relationships between consumers and producers of digital content will place strain on transport protocols such as TCP, as those protocols are not designed to accomodate the fluid interactions which will take place when users have a multiplicity of service options. The researcher argues for lightweight reliable transport services designed for maximum flexibility which can accomodate connection preemption, connection suspension, connection migration and connection aggregation. While these considerations are fundamental , they have not been carefully studied in the context of TCP because of daunting technical hurdles, complexity, and absence of a driving need. The researcher brings new techniques to bear on the problem, as the proposed work leverages his experience with fast forward error correcting coding techniques which he has applied in the domain of reliable multicast [11] and downloading digital content from multiple mirror sites in parallel [10]. In some of the settings considered, such as dynamic server selection and redirection, the flexibility which the proposed techniques afford can be considered to be a performance optimization, but in other settings, such as providing reliability for end system multicast, the researcher argues that the flexible services proposed are essential to delivering high performance. The proposed educational aspect of the researcher's career development program focuses on enhancing the undergraduate and graduate student experience in his department and building a stronger basis for industrial and inter-institutional networking research collaborations in the Boston area. Proposed activities center around ongoing projects to promote undergraduate involvement in networking research; to enhance the department's networking curriculum; and to encourage students to participate in the computer science community outside of the university, whether by volunteering at a community outreach program or by becoming involved in local organizations of computer scientists. An exciting initiative which the researcher is e actively involved in and which will help achieve these goals is the institution of a center for networking research at Boston University.