Energy Efficient Policies for Upstream Content Server Selection

摘要

Reverse proxies are key performance elements of a content distribution network (CDN) and operate as intermediaries between client and upstream (or origin) server traffic. A typical reverse proxy server aims to distribute user workload, more or less equally, among multiple upstream servers. To keep the pace with the growing demand for content, CDNs regularly expand their capacity by adding extra upstream servers to handle the increased traffic. However, distributing traffic without consideration of the energy and performance features of the machines and keeping servers active all the time can reduce energy efficiency drastically because of the seasonal variation of traffic. This outcome is highly undesirable because of environmental and economic concerns. Policy-based management is an approach for large-scale network management that helps to simplify network administration through the use of event-action rules. We investigate the formulation of server selection policies for reverse proxies to improve the energy efficiency of the system while maintaining desired service-level objectives. In particular, we formulate four energy-efficient policies of varying complexity and performance, and evaluate them using realistic data obtained from a testbed. The results are also contrasted with the outcome of common load distribution methods that ship with a popular open-source proxy server to show the advantages of the proposed policy generation strategies.