Cloud computing Infrastructure-as-a-Service (IaaS) has gained momentum in the cloud computing research field due to its ability to provide efficient infrastructures. Cloud Service Providers (CSPs) are striving to offer Quality of Service (QoS)-guaranteed IaaS services while also improving their resource utilization and maximizing profit. In addition, CSPs are challenged by the need to manipulate diverse and heterogeneous resources, realizing multiple objectives for both customers and CSPs, and handling scalability issues. These challenges are the motivations behind this work which aims at developing a multi-layered framework for constructing and managing efficient IaaS. The fundamental layer in this
framework, the Virtual Infrastructure (VI) composition layer, is dedicated to composing and delivering VIs as an IaaS service. This framework relies on a preparatory step that is defined when all the available resources in the managed space are collected in a large repository, the Virtual Resource Pool (VRP). The VRP creation process unifies the representation of all the diverse and heterogeneous resources available.
Subsequently, the proposed framework performs various resource allocation approaches as working solutions through the VI composition layer. These approaches adopt efficient techniques and methodologies in performing their operations. The working solutions are initiated by designing a composition approach that relies on an ontology-based model representation. The composition approach exploits semantic similarity, closeness centrality, and random walk techniques for efficient resource allocation. As a result, it provides an efficient solution in a reasonable computational time with no guarantee for the optimality of the
obtained solutions. To achieve an optimal solution, the composition approach uses a mathematical modeling formulation. In this solution, the concepts of the composition approach have been integrated into a multi-objective Mixed Integer Linear Programming (MILP) model that has been solved optimally. Despite the optimality of the resulting solution, the MILP-based model restricts IaaS resource allocation to a computational running-time challenge, and the issue of limited-size datacenters. To circumvent these issues, a cost-efficient model is proposed. The new model introduces a Column Generation (CG) formulation for the IaaS resource allocation problem in large datacenters acquainted with QoS requirements. Furthermore, this formulation is realistic, adopts large-scale optimization tools that are adequate for large datacenters, and ensures optimal solutions in a reasonable time.
However, growing costs in large datacenters in accordance with the growth of recent
large-scale application demands, makes large datacenters economically inefficient. Thus, we advocate a distributed framework for IaaS provisioning that guarantees affordable, scalable, and QoS-assured infrastructure for hosting large-scale applications in geo-distributed datacenters. The framework incorporates two decentralized resource allocation approaches, hierarchical and distributed, that use efficient economic models. These approaches are quite promising solutions for the scalability and computational complexity issues of existing centralized approaches. Finally, the cost-efficient model has been extended to fit the distributed infrastructure by considering additional constraints that impact CSP revenue. Simulation results showcase the effectiveness of the presented work along with the potential benefits of the proposed solutions in terms of satisfying the customers’ requirements, while
achieving a better resource utilization and CSP payoffs.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/34951 |
| Date | January 2016 |
| Creators | Metwally, Khaled |
| Contributors | Karmouch, Ahmed |
| Publisher | Université d'Ottawa / University of Ottawa |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0022 seconds