Enhancing quality of service in cloud computing through novel resource management

Distributed Systems as an area of research has seen a gradual evolution over the last few decades fuelled by the application of new use cases to technological developments.
Cloud Computing is one such paradigm that has evolved from the adoption of Utility Computing, Virtualization and Service Oriented Architectures. Cloud Computing can
be distinguished from other distributed paradigms though the provisioning of resources, data and software to users on demand in a similar fashion to the services provided by the electric power industry. Commercial Cloud offerings are expected to meet the Quality of Service (QoS) requirements of a consumer via Service Level Agreements (SLA). In reality, Cloud providers rarely provide QoS beyond best effort as the intrinsic fault tolerant nature of currently deployed applications require little more. Nevertheless, with enhancements to QoS in Cloud Computing the range of deployable applications can be improved and thus advance the overall adoption of the paradigm. This thesis tackles the shortcoming of QoS in Cloud Computing though novel enhancements to Cloud resource management. Since QoS is a broad subject area, the scope of research within has been narrowed down to two specific areas of interest: performance
and scalability. In this thesis, the performance and scalability of Cloud technology are ascertained through performance evaluations on Hypervisor (such as XEN and KVM) and Cloud Infrastructure Managers (such as OpenNebula and Nimbus). Recommendations are made on how to resolve performance bottlenecks and on the suitability of certain
technology for specific Cloud applications. Contextualisation and Re-contextualization mechanisms are introduced for self-configuring virtual Cloud resources at operation time while managing resources and software dependencies at the infrastructure and platform layer of the Cloud software stack. In addition, the thesis aims to improve the adoption of the Cloud by exploring novel techniques for composing, configuring and deploying Grid
Middleware onto Cloud resources. The core contributions of this thesis are as follows: i) A prototype software tool for
the (re-)contextualization of Cloud applications, platforms, infrastructures and resource dependencies that enables improvements to performance, scalability and fault tolerance.
ii) Performance results and recommendations on the topic of Virtual Machine (VM) image propagation delay in Cloud infrastructure technology, Paravirtualized block device
drivers and VM image standards in Hypervisor technology, for the purpose of ascertaining current limitations in Cloud QoS. iii) A software prototype system of an interoperable self-configuring Virtual Grid infrastructure, deployable on to a range of Cloud providers, to enhance the QoS achievable by Grid applications.