Dynamic server selection using fuzzy inference in content distribution networks

Abstract

To accommodate the exponential growth of Web traffic, Content Distribution Networks (CDN) have been designed and deployed to distribute content to different cache servers, and to transparently and dynamically redirect user requests to the cache servers according to the latest network and server status. Server selection therefore is vital and crucial to both the functionality and performance of any CDN systems. An appropriate server should be selected by taking estimated user location, measured round-trip time, and advertised server load into account. However, it is unlikely to obtain accurate and timely inputs of these parameters in practice, so that the effectiveness and efficiency of CDN cannot be fully achieved by traditional means. In this paper, a novel CDN server selection scheme using fuzzy inference is proposed. The scheme selects appropriate servers based on partial round-trip time measurements and historical server load information, and it can be implemented generically wherever the decision is made. It is shown that the fuzzy inference-based scheme is inherently capable of handling multiple decision inputs efficiently, tolerable to measurement noise and errors, and able to deal with network dynamics. Simulation results demonstrate that, compared with other server selection schemes, the proposed scheme can achieve higher resource utilization, provide better user-perceived Quality of Service (QoS), and efficiently deal with network dynamics.