| The widespread adoption of cloud computing makes cloud services a core component of modern IT systems. Cloud service monitoring brings benefits to both service providers and consumers. It enables customers to track the status and quality of their services to ensure that their SLAs are satisfied and possibly record any SLA violations. It also supports service providers to predict potential violations and deploy additional resources in advance to avoid possible penalties of performance violations. However, federated cloud poses challenges to application monitoring due to the lack of cross-platform interoperability, management, and vendor lock-in constraints. In addition, SLA definitions might be different from one service provider to another.;Many cloud providers offer tools to enable customers to monitor the performance of their cloud-based services. The major drawback of these commercial monitoring tools is that they do not present enough detail to consumers to be able to customize the behavior of these tools to accommodate their specific requirements. In addition, such monitoring tools typically are provider-dependent, which means that consumers cannot use the same monitoring platform for multiple cloud providers.;In this thesis, we present SLAM, a flexible framework for SLA monitoring in federated cloud environments. SLAM is platform-independent, which enables interoperability between different cloud providers. SLAM can monitor physical and virtual resources. It uses user-defined rules and policies to map SLA parameters to relevant low-level metrics. SLAM supports monitoring of distributed nodes and hosts using an agent-based model. It supports different monitoring frequencies (i.e. checking rate) for each monitoring metric. In addition, we propose a service benchmarking approach that can compare similar services offered by different cloud providers without deploying monitoring agents at the provider's site. We implemented a poof-of-concept prototype to showcase the flexibility and usability of our proposed system. Performance evaluation shows that our proposed system is scalable and poses a very limited overhead on target systems. |
