Vageesan, Gopinath (2005-12). On improving performance and conserving power in cluster-based web servers. Master's Thesis.
Efficiency and power conservation are critical issues in the design of cluster systems
because these two parameters have direct implications on the user experience and the
global need to conserve power. Widely adopted, distributor-based systems forward client
requests to a balanced set of waiting servers in complete transparency to the clients. The
policy employed in forwarding requests from the front-end distributor to the backend
servers plays an important role in the overall system performance. Existing research
separately addresses server performance and power conservation. The locality-aware
request distribution (LARD) scheme improves the system response time by having the
requests served by web servers which have the data in their cache. The power-aware
request distribution aims at reducing the power consumption by turning the web servers
OFF and ON according to the load.
This research tries to achieve power conservation while preserving the performance of the
system. First, we prove that using both power-aware and locality-aware request
distribution together provides optimum power conservation, while still maintaining the
required QoS of the system. We apply the usage of pinned memory in the backend servers to boost performance along with a request distributor design based on power and locality
considerations. Secondly, we employ an intelligent-proactive-distribution policy at the
front-end to improve the distribution scheme and complementary pre-fetching at the backend
server nodes. The proactive distribution depends on both online and offline analysis of
the website log files, which capture user navigation patterns on the website. The prefetching
scheme pre-fetches the web pages into the memory based on a confidence value of
the web page predicted by backend using the log file analysis. Designed to work with the
prevailing web technologies, such as HTTP 1.1, our scheme provides reduced response
time to the clients and improved power conservation at the backend server cluster.
Simulations carried out with traces derived from the log files of real web servers witness
performance boost of 15-45% and 10-40% power conservation in comparison to the
existing distribution policies.
Kim, Eun Associate Professor - Term Appoint