Queen's School of Computing

PhD Candidate: A B MBodrul Alam

Supervisor: MohammadZulkernine, Professor, Queen’s Computing

External Examiner:Abdallah Shami, Professor, ECE, Western University

Internal/ExternalExaminer: Aboelmagd Noureldin, RMC, ECE Queen’s (cross-appointed), Computing(cross-appointed)

Internal Examiner:Patrick Martin, Professor, Queen’s Computing

Head’s Rep: HossamHassanein, Professor, Queen’s Computing

Date and time: August 17,10:00am

Title: TowardsReliability Evaluation and Integration in Cloud Resource Management


Theadoption of cloud computing technology is increasing day by day due to itsenormous services and low cost. Customers transfer their businesses to cloudbecause of the beneficial cloud features such as scalability, high performance,on-demand, and pay-per-use service model. However, transferring services tocloud adds a new level of risk due to loss of control, which makes servicereliability an essential driving factor of the cloud market today, especiallyin light of the recent cloud failures and outages that raise customers'concerns.

In thisthesis, the goal is to propose a reliability evaluation model and integrate thedeveloped model in cloud resource management such as Virtual Machine (VM)allocation, VM migration, and cloud federation formation in order to increasethe cloud service reliability. In this thesis, firstly a cloud reliabilityevaluation model is proposed. Some types of failures from different domains ofthe cloud environment are considered to evaluate cloud reliability. To proposethe evaluation model, a classification strategy for cloud failures is alsooutlined. Secondly, to show the impact of the integration of the proposedmodel, a multi-objective placement model for interdependent VMs in cloud isproposed while considering both reliability and Quality of Service (QoS). Amulti-objective genetic algorithm is used to solve the placement problemheuristically. Thirdly, a Markov-based failure prediction model is proposed toanticipate the failures of cloud servers. The proposed prediction model is thenintegrated into a VM migration model in a multi-cloud setting to maximize cloudreliability while reducing VM communication delay. The VM migration problem issolved optimally and heuristically using the Artificial Bee Colony (ABC)algorithm. Finally, a reliability-based cloud federation model is proposed usinga hedonic coalition formation game based on a reliability-driven utilityfunction. All proposed models will serve as a guide to both customers and cloudservice providers towards the achievement of reliable resource allocation,migration, and federation formation. The effectiveness of all the models hasbeen demonstrated through experiments. The evaluation shows that the models arecomputationally efficient and achieve high cloud reliability.