Reliability Assessment for Cloud Applications

Reliability is a significant quality measurement for computer systems and applications. When cloud computing is becoming mature and pervasive, a variety of applications are deployed on cloud platforms. Influenced by the characteristics of cloud computing, such as server consolidation and virtualization, and reliability requirements, such as redundancies, the deployment structure of cloud applications are complex. A deployment structure usually involves components like services, Virtual Machines (VMs), Physical Servers (PSs), etc. Dependencies of these components make the reliability assessment of cloud applications challenging. In this thesis, we intend to model cloud applications based on their deployment structures and give an accurate reliability assessment method. To this aim, we propose a DEpendency-Based Reliability Assessment (DEBRA) and accordingly design a framework composed of three functional components: a dependency analyzer for analyzing dependencies between components used by applications and modeling cloud applications with Layered Dependency Graphs (LDGs), a monitor for deriving states of application components, and a reliability analyzer based on DEBRA for assessing the reliability of components as well as applications. Furthermore, we implement the three components and two extra functional components for usage-based testing, which are a fault injector for injecting failures to application components and a tester for testing applications. We apply our framework to a real-world application and cloud platforms, and conduct case studies. In these case studies, we verify if our framework can assess the cloud application reliability accurately and precisely. We deploy the application with several different structures to a cloud and test the application with usage-based requests. We then use DEBRA and several existing methods to assess the application reliability. The assessment results are compared according to a proposed comparison process. The comparison results show that DEBRA can obtain results of high quality and has several merits regarding modeling cloud applications for reliability assessment.