Cloud Service Level planning under burstiness

Cloud Service Providers (SPs) need tools to help them plan their infrastructure capacity and decide on Service Level Agreements (SLAs) with their customers prior to deploying customer's applications. Existing approaches have not considered several important challenges such as workload burstiness, workload uncertainty, and scalability to large number of applications. This paper proposes a trace-based framework to address these challenges simultaneously. The core of the framework is a novel Resource Allocation Planning (RAP) methodology which allows SPs to take into account workload burstiness in Service Level Planning (SLP). This methodology works in consort with a Monte Carlo simulation technique to systematically consider the impact of workload uncertainty in SLP. Furthermore, we propose a novel burstiness-aware clustering technique that groups applications with similar workload characteristics to improve the scalability of the SLP framework. Results show that our approach can yield near optimal resource allocations and can achieve lower SLO violations with fewer resources than competing approaches, especially for bursty workloads. Furthermore, our proposed clustering technique is able to improve the scalability of our SLP framework without significantly impacting resource allocation accuracy.