Abstract

To meet user requirements, systems can be built from Commercial-Off-The-Shelf (COTS) components, potentially from different vendors. However, the gap between the requirements referring to the overall system and the components to build the system from can be large. To close the gap, it is required to decompose the requirements to a level where they can be mapped to components.
When the designed system is deployed and ready for operations, its services are sold and pro-vided to customers. One important goal for service providers is to optimize system resource utilization while ensuring the quality of service expressed in the Service Level Agreements (SLAs). For this purpose, the system can be reconfigured dynamically according to the cur-rent workload to satisfy the SLAs while using only necessary resources. To manage the re-configuration of the system at runtime, a set of previously defined patterns called elasticity rules can be used. In elasticity rules, the actions that need to be taken to reconfigure the sys-tem are specified. An elasticity rule is generally invoked by a trigger, which is generated in reaction to a monitoring event.
In this thesis, we propose a model-driven management framework which aims at user re-quirements satisfaction, SLA compliance management and enabling dynamic reconfiguration by reusing the design information at runtime.
An approach has been developed to derive automatically a valid configuration starting from low level requirements called service configurations. However, the service configurations are far from requirements a user would express. To generate a system configuration from user requirements and alleviate the work of designer, we generate service configurations by de-composing functional user requirements to the level where components can be selected and put together to satisfy the user requirements. We integrated our service configurations gen-erator with the previous configuration generator.
In our framework, we reuse the information acquired from system configuration and dimen-sioning to generate elasticity rules offline. We propose a model driven approach to check the compliance of SLAs and generate triggers for invoking applicable elasticity rules when system reconfiguration is required. For handling multiple triggers generated at the same time, we propose a solution to automatically correlate the actions of invoked elasticity rules, when re-quired. The framework consists of a number of metamodels and a set of model transfor-mations. We use the Unified Modeling Language (UML) and its profiling mechanism to de-scribe all the artifacts in the proposed framework. We implement the profiles using Eclipse Modeling Framework (EMF) and Papyrus. To implement the processes, we use the Atlas Transformation Language (ATL). We also use the APIs of the Object Constraint Language (OCL) in the Eclipse environment to develop a tool for checking constraints and generating triggers.