Feedback-based Scheduling for Back-end Databases in Shared Dynamic Content Server Clusters

This paper introduces a self-configuring architecture for scaling the database tier of dynamic content Web servers. We use a unified approach to load and fault management based on dynamic data replication and feedback-based scheduling. While replication provides scaling and high availability, feedback scheduling dynamically allocates tasks to commodity databases across workloads in response to peak loads or failure conditions thus providing quality of service. By augmenting the feedback loop with state awareness, we avoid oscillations in resource allocation. We investigate our transparent provisioning mechanisms in the database tier using the TPC-W e-commerce and the on-line auction Rubis benchmarks. We demonstrate that our techniques provide quality of service under load bursts and failure scenarios