Data Stream Splicing for Web Proxy Cache Optimization

Redundant data copies in Web proxy cache applications result in poor performance. This paper analyzes the data stream movement of Squid, which is a widely used web proxy cache program. Data stream splicing optimization to Squid includes three models: splicing client and server socket, splicing local file cache and client socket, splicing server socket and local file cache. All these splicing operations are implemented in Linux kernel module and corresponding system call interfaces are provided for applications to use. Data stream splicing not only eliminates data copies between user space and kernel space, but also eliminates the translations of user virtual address to physical address. Experiments prove that data stream splicing can reduce mean response time and data transfer overheads, and improve the throughput of proxy servers remarkably most of the time