Instantaneous Offloading of Web Server Load

by Vsevolod V. Panteleenko Traffic to a web server is unpredictable and very peaky. Such peaks can be up to ten times the average load. To avoid losing or alienating clients, web servers are overbuilt to handle the peak load. Consequently, there are huge investments in local infrastructure at web sites that are significantly underutilized except for infrequent peak periods. This dissertation describes the design and performance study of a web booster architecture, which offloads a web server during peak traffic. The architecture instantaneously decreases the processing cost on a web server for requests to both static and dynamic web resources. It consists of a booster appliance, which pre-processes client requests, and an accelerator software module running on the web server. Fast reaction to the load change and releasing unused resources during inactive period allows multiple web servers to efficiently share the booster appliance. The main contributions of this dissertation are: (i) a performance analysis of two web servers using a simulated WAN workload, which indicates that the main cost of request processing is related to network protocol processing; (ii) a design, implementation and analysis of web server offloading techniques that include request pre-processing by the booster and a packet cache implemented by the accelerator module, which together decrease the server load up to an order of magnitude; and (iii) a design, implementation and analysis of the booster appliance and techniques for improving its performance. Vsevolod Vasiliovich Panteleenko