Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Computer Science


Hanan Lutfiyya


In the last decade, Cloud Computing has become a disruptive force in the computing landscape, changing the way in which software is designed, deployed and used over the world. Its adoption has been substantial and it is only expected to continue growing. The growth of this new model is supported by the proliferation of large-scale data centres, built for the express purpose of hosting cloud workloads. These data centres rely on systems virtualization to host multiple workloads per physical server, thus increasing their infrastructures' utilization and decreasing their power consumption. However, the owners of the cloud workloads expect their applications' demand to be satisfied at all times, and placing too many workloads in one physical server can risk meeting those service expectations. These and other management goals make the task of managing a cloud-supporting data centre a complex challenge, but one that needs to be addressed.

In this work, we address a few of the management challenges associated with dynamic resource management in virtualized data centres. We investigate the application of First Fit heuristics to the Virtual Machine Relocation problem (that is, the problem of migrating VMs away from stressed or overloaded hosts) and the effect that different heuristics have, as reflected in the performance metrics of the data centre. We also investigate how to pursue multiple goals in data centre management and propose a method to achieve precisely that by dynamically switching management strategies at runtime according to data centre state. In order to improve system scalability and decrease network management overhead, we propose architecting the management system as a topology-aware hierarchy of managing elements, which limits the flow of management data across the data centre. Finally, we address the challenge of managing multi-VM applications with placement constraints in data centres, while still trying to achieve high levels of resource utilization and client satisfaction.