Systematically Translating Service Level Objectives into Design and Operational Policies for Multi-Tier Applications

A Service Level Agreement (SLA) contains one or more Service Level Objectives (SLOs) that describe the agreed upon quality requirements at the service-level. In order to manage the service to meet the agreed upon SLA, it is important not only to design a service of the required capacity but also to monitor the service thereafter for violations at runtime. This objective can be achieved by undertaking SLA Decomposition, i.e., translating SLOs specified in the SLA into lower-level policies that can then be used for design and enforcement purposes. Such design and operational policies are often constraints on thresholds of lower level metrics. Traditionally, domain experts and administrators bring their knowledge to bear upon the problem of SLA decomposition. This practice is ad-hoc, manual, and static (i.e., done once). This is both costly, and not well suited to dynamic workloads. In the past, there has been a number of efforts to develop more automated and dynamic solutions, but these approaches have many limitations and hence pose major challenges to their applicability in practice. In this paper, we propose a systematic and practical approach that combines fine-grained performance modeling with regression analysis to translate service level objectives directly into design and operational policies for multi-tier applications. We demonstrate that our approach is practical and can be applied to commonly used multi-tier applications with different topologies and performance characteristics. Our approach handles both request-based and session-based workloads and deals with workload changes in terms of both request volume and transaction mix. Our approach is non-intrusive in the sense that it requires no specialized profiling, i.e., the data used in our approach is readily available from normal system and application monitoring. We validate our approach using both the RUBis e-commerce benchmark and a trace-driven simulation of a business-critical enterprise application. These results show the effectiveness of our approach.