Service Diferentiation and Overload Control in Internet Servers

Michigan State University, East Lansing MI

Investigators

Abstract

The increasing volume and evolving types of Internet applications have been demanding enhanced service from the Internet infrastructure. A significant amount of work is in progress on the networking aspects of the Internet. To complement these efforts, the Principal Investigator (PI) proposes to explore the issues at the server level. Specifically, this proposal addresses the support for service differentiation and overload control in Internet servers. Support for service differentiation at the servers is necessary because of two primary reasons. First, evolving applications like e-commerce and continuous media transmission will be highly benefited by the differentiated service model. Second, the adoption (most likely) of some form of service differentiation (at the router level) in the next-generation Internet will demand the servers to reciprocate in their services. The proposed research in this context will be approached in two steps: (a) formulation of algorithms for service differentiation, and (b) evaluation of the effectiveness and performance of service differentiating Internet servers. Overloads at the servers are unavoidable because of the bursty and unpredictable nature of the Internet traffic. Overload causes severe performance degradation or even crashes the server. E-commerce businesses incur serious financial losses because of the overloaded servers. The performance degradation of overloaded servers may be greatly reduced by incorporating overload control mechanisms at the servers. The PI proposes to address the design and implementation of overload control mechanisms at the application level. In general, the performance and functionality of the Internet servers will be greatly enhanced by the adoption of service differentiation and overload control mechanisms. The outcome of the proposed research will have a broader impact on the Internet infrastructure and on its users. Specifically, the proposed techniques will be complementary to the efforts in the direction of the next-generation Internet, and will be helpful for the e-commerce industry. They will also facilitate the support of continuous media objects in the Internet. At the local level, the research will have a positive educational impact on the students of Michigan State University.