| Cloud computing is being widely accepted and utilized in the business world. From the perspective of businesses utilizing the cloud, it is critical to meet their customers' requirements by achieving service-level-objectives. Hence, the ability to accurately characterize and optimize cloud-service performance is of great importance.;In this dissertation, a stochastic multi-tenant framework is proposed to model the service of customer requests in a cloud infrastructure composed of heterogeneous virtual machines (VMs). The proposed framework addresses the critical concepts and characteristics in the cloud, including virtualization, multi-tenancy, heterogeneity of VMs, VM isolation for the purpose of security and/or performance guarantee and the stochastic response time of a customer request. Two cloud-service performance metrics are mathematically characterized, namely the percentile of the stochastic response time and the mean of the stochastic response time of a customer request.;Based upon the proposed multi-tenant framework, a workload-allocation algorithm, termed max-min-cloud algorithm, is then devised to optimize the performance of the cloud service. A rigorous optimality proof of the max-min-cloud algorithm is given when the stochastic response time of a customer request assumed exponentially distributed. Furthermore, extensive Monte-Carlo simulations are conducted to validate the optimality of the max-min-cloud algorithm by comparing with other two workload-allocation algorithms under various scenarios.;Next, the resource provisioning problem in the cloud is studied in light of the max-min-cloud algorithm. In particular, an efficient resource-provisioning strategy, termed the MPC strategy, is proposed for serving dynamically arriving customer requests. The efficacy of the MPC strategy is verified through two practical cases when the arrival of the customer requests is predictable and unpredictable, respectively.;As an extension of the max-min-cloud algorithm, we further devise the max-load-first algorithm to deal with the VM placement problem in the cloud. MC simulation results show that the max-load-first VM-placement algorithm outperforms the other two heuristic algorithms in terms of reducing the mean of stochastic completion time of a group of arbitrary customers' requests. Simulation results also provide insight on how the initial loads of servers affect the performance of the cloud system.;In summary, the findings in this dissertation work can be of great benefit to both service providers (namely business owners) and cloud providers. For business owners, the max-min-cloud workload-allocation algorithm and the MPC resource-provisioning strategy together can be used help them build a better understanding of how much virtual resources in the cloud they may need to meet customers' expectations subject to cost constraints. For cloud providers, the max-load-first VM-placement algorithm can be used to optimize the computational performance of the service by appropriately utilizing the physical machines and efficiently placing the VMs in their cloud infrastructures. |
