Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 100 === Today''s data centers face serious challenges in providing low-latency high-quality services to meet user''s expectations for interactivity. In such environment, a network flow is useful, and contributes to application-level throughput if and only if it is delivered before its deadline.Therefore, in recently years,some researchers have employed application-level deadline information for satisfying the latency requirement. A fundamental question arises: does high application-level throughput mean providing high-quality service? We investigate behaviour of users and the content of the flow, which consists of many responses to the query. We found that each flow may contribute a few highly correlative responses to the query and the rest are less-important responses. This observation motivates us to associate the importance of the flow contents with the application traffic for further improving the service quality.We model the application importance maximization problem in a generic and a server-centric network. Because both of them are too complex, we propose the importance-aware delivery protocol, which is a novel distributed event-driven rate-based delivery control protocol, for server-centric data center networks. The proposed protocol utilizes the multiple disjoin paths of server-centric network, and jointly considers flow importance, flow size, and deadline to maximize the importance contribution of flows in the network. Through real-data-based or synthetic simulations, we demonstrate that the proposed protocol significantly outperforms deadline-aware protocol and MPTCP in terms of the precision at K and the sum of application-level importance.
|