Elastic and Real-time Capacity Planning for Web Search Engines

Abstract

Web search engines are complex systems deployedon large clusters of processors. With a constant growing numberof users, Web search engines capacity must be frequentlyevaluated. Among other techniques, capacity planning methodsare a good approach to estimate the amount of processors incluster based systems, allowing data-center engineers to ensureenough computational resources are available to efficientlydeal with the ever changing streams of user queries. In thispaper, we present a novel capacity planning methodology thatcombines discrete-event simulation and classical operationalanalysis for open systems formulae. With this methodology,just a small fraction of deployments are evaluated to find anoptimal assignment of resources given a target workload. Weexperimentally evaluate a simulation-based capacity planningmodule in combination with an OpenMP+MPI implementationof a Web search engine, that periodically evaluates itsperformance metrics in order to elastically adjust the amountof computational resources needed to support the currentworkload. Results show that the proposed methodology isable to quickly provide a new deployment configuration uponperformance changes, ensuring that a strictly sufficient amountof computational resources are available to efficiently deal withusers demands.