Towards effective live cloud migration on public cloud IaaS.

Abstract

Cloud computing allows users to access shared, online computing resources. However, providers often offer their own proprietary applications, APIs and infrastructures, resulting in a heterogeneous cloud environment. This environment makes it difficult for users to change cloud service providers and to explore capabilities to support the automated migration from one provider to another. Many standards bodies (IEEE, NIST, DMTF and SNIA), industry (middleware) and academia have been pursuing standards and approaches to reduce the impact of vendor lock-in. Cloud providers offer their Infrastructure as a Service (IaaS) based on virtualization to enable multi-tenant and isolated environments for users. Because, each provider has its own proprietary virtual machine (VM) manager, called the hypervisor, VMs are usually tightly coupled to the underlying hardware, thus hindering live migration of VMs to different providers. A number of user-centric approaches have been proposed from both academia and industry to solve this coupling issue. However, these approaches suffer limitations in terms of flexibility (decoupling VMs from underlying hardware), performance (migration downtime) and security (secure live migration). These limitations are identified using our live cloud migration criteria which are rep- resented by flexibility, performance and security. These criteria are not only used to point out the gap in the previous approaches, but are also used to design our live cloud migration approach, LivCloud. This approach aims to live migration of VMs across various cloud IaaS with minimal migration downtime, with no extra cost and without user’s intervention and awareness. This aim has been achieved by addressing different gaps identified in the three criteria: the flexibility gap is improved by considering a better virtualization platform to support a wider hardware range, supporting various operating system and taking into account the migrated VMs’ hardware specifications and layout; the performance gap is enhanced by improving the network connectivity, providing extra resources required by the migrated VMs during the migration and predicting any potential failure to roll back the system to its initial state if required; finally, the security gap is clearly tackled by protecting the migration channel using encryption and authentication. This thesis presents: (i) A clear identification of the key challenges and factors to successfully perform live migration of VMs across different cloud IaaS. This has resulted in a rigorous comparative analysis of the literature on live migration of VMs at the cloud IaaS based on our live cloud migration criteria; (ii) A rigorous analysis to distil the limitations of existing live cloud migration approaches and how to design efficient live cloud migration using up-to-date technologies. This has led to design a novel live cloud migration approach, called LivCloud, that overcomes key limitations in currently available approaches, is designed into two stages, the basic design stage and the enhancement of the basic design stage; (iii) A systematic approach to assess LivCloud on different public cloud IaaS. This has been achieved by using a combination of up-to-date technologies to build LivCloud taking the interoperability challenge into account, implementing and discussing the results of the basic design stage on Amazon IaaS, and implementing both stages of the approach on Packet bare metal cloud. To sum up, the thesis introduces a live cloud migration approach that is systematically designed and evaluated on uncontrolled environments, Amazon and Packet bare metal. In contrast to other approaches, it clearly highlights how to perform and secure the migration between our local network and the mentioned environments.