Self Tuning for Elastic Storage in Cloud Environment

Abstract: Elasticity, where a system requests and releases resources in response to a dynamic property, has been an important issue in Cloud computing. It can be handled manually or automatically. Efforts being made to make elasticity as automatic as possible. Autonomic computing has played a significant role in many computing fields including Cloud computing. In this master thesis, we have adopted control theory approach for automation of elasticity in key-value storage that is provided in a cloud environment and operates under dynamic workloads. Automation is achieved by providing a feedback controller that automatically grows and shrinks the number of nodes in order to meet Service Level Agreement (SLAs) under high load and reduces costs under low load. Every step of building a controller for elastic storage, including System Identification and controller design, is discussed in this thesis. We have evaluated our approach by simulation. We have implemented a simulation framework based on Kompics1, in order to simulate an elastic key-value store in Cloud environment and to be able to experiment with different controllers. Finally, we have examined the implemented controller against specific SLA requirements and we have evaluated the controller behaviors in different scenarios. Our simulation experiments have shown the feasibility of our approach to automate elasticity of storage services.