servload: Generating representative workloads for web server benchmarking

Web server benchmarking has two main purposes, comparing different hardware platforms or server configurations, and testing the scalability of a given system. For the first purpose, the capability to replay given workload traces is needed. For the second purpose, also the capability to generate increased representative workloads is needed. Scalability tests are of special interests for Internet Service Providers (ISPs) to estimate performance numbers for Service Level Agreements (SLAs) contracted to customers. Depending on the setup and the implemented web application the required scalability may vary from a few requests per seconds up to a waste amount of requests. This paper presents a web server benchmark called servload which generates modified workloads from given traces keeping similar characteristics. servload is generic in the sense that it is not dedicated to a special application field. It can process traces from different application fields for example banking, e-commerce or collaborative wikis. The different workload generation methods of servload are discussed and evaluated. In the presented experiments servload shows good replay capabilities.