Web Booster Architecture for Accelerating Web Servers

This paper describes a novel approach to increase the performance of web sites by changing the network traffic between the client and the web servers. This is achieved by introducing a web booster architecture, which 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 the release of resources during inactive period enables multiple web servers to efficiently time share the booster appliance. The paper describes six techniques that decrease cost of the server processing up to nine times. These techniques are: (i) increasing MTU size of the connections to a web server, (ii) avoiding connection open and close, (iii) delaying client acknowledgements, (iv) performing request processing in the kernel, (v) avoiding data copying and checksumming, and (vi) avoiding packet creation overhead. The paper describes a web accelerator software module running on the web server that enables the last three of the above optimizations by caching the requested documents in the kernel in the form of TCP packets. The paper also presents a design of the booster appliance and four novel optimizations of its network protocol stack that increase booster performance up to a factor of four.