DBSeer: Resource and Performance Prediction for Building a Next Generation Database Cloud

Cloud computing is characterized by shared infrastructure and a decoupling between its operators and tenants. These two characteristics impose new challenges to databases applications hosted in the cloud, namely: (i) how to price database services, (ii) how to isolate database tenants, and (iii) how to optimize database performance on this shared infrastructure. We argue that today’s solutions, based on virtual-machines, do not properly address these challenges. We hint at new research directions to tackle these problems and argue that these three challenges share a common need for accurate predictive models of performance and resource utilization. We present our approach, called DBSeer, with our initial results on predictive models for the important class of OLTP/Web workloads and show how they can be used to address these challenges.