Abstract

Cloud computing is becoming popular in the computing industry. Elasticity and availability are two features often associated with cloud computing. Elasticity is defined as the automatic provisioning of resources when needed and de-provisioning of resources when they are not needed. Cloud computing offers the users the option of only paying for what they use and guarantees the availability of virtual infrastructure (i.e. virtual machines). The existing cloud solutions handle both elasticity and availability at the virtual infrastructure level through the manipulation, restart, addition and removal of virtual machines (VMs) as required. These solutions equate the application and its workload to the VMs that run the application. High-availability applications are typically composed of redundant resources, and recover from failures through failover mostly managed by a middleware. For such applications, handling elasticity at the virtual infrastructure level through the addition and removal of VMs is not enough, as the availability management in application level will not make use of additional resources. This requires new solutions that manage both elasticity and availability in application level. In this thesis, we provide a solution to manage the elasticity and availability of applications based on a standard middleware defined by the Service Availability Forum (SA Forum). Our solution manages application level elasticity through the manipulation of the application configuration used by the middleware to ensure service availability. For this purpose we introduce a third party, ‘Elasticity Engine’ (EE), that manipulates the application configuration used by the SA Forum middleware when a workload changes. This in turn triggers the SA Forum middleware to change the workload distribution in the system while ensuring service availability. We explore the SA Forum middleware configuration attributes that play a role in elasticity management, the constraints applicable to them, as well as their impact on the load distribution. We propose an overall architecture for availability and elasticity management for an SA Forum system. We design the EE architecture and behavior through a set of strategies for elasticity. The proposed EE has been implemented and tested.