Research On Cost-Aware Virtual Resource Provisioning Mechanisms For Cloud Services

As an emerging provisioning paradigm of IT resources, cloud computing has developed very quickly in the recent years. Cloud computing aims at providing higher quality elastic cloud services at a low cost to users on demand, and the operation status of cloud services mainly depends on the fact that whether the virtual resources needed by cloud services can be provisioned cost-effectively.With the increasing size of the virtualized data center, enhanced heterogeneity, and diversification of cloud service application type and their resource demand characters, high energy consumption, dynamic resource management as well as the contradiction between application performance and resource provisioning cost become more prominent than ever. In order to solve these problems and contradictions, a lot of work has been conducted, and much has been achieved. However, much has remains to be done in terms of the energy-aware resource provisioning for IaaS cloud services, dynamic multi-objective resource provisioning optimization of large-scale virtualized data center and SaaS cloud service request scheduling. Therefore, centered on the cost and profit of infrastructure operator (e.g., IaaS providers) and cloud service providers (e.g., SaaS providers) respectively, we have conducted a detailed research on the cost-aware virtual resource provisioning mechanisms for IaaS/SaaS cloud service, and achieved the following results.1. In order to solve the problem of high energy consumption of large-scale virtualized data centers, an energy-aware virtual machine static placement intelligent optimization algorithm based on improved discrete particle swarm optimization is presented. According to the essence of the virtualized data center energy consumption optimization, the proposed algorithm models the energy consumption of heterogeneous virtualized data center as a multiple dimension bin-packing problem with variable sizes, and redefines the parameters and operators of particles based on the characters of virtual machine placement problem. Furthermore, a novel two dimension particle encoding method is presented. A fitness proportion method based on adaptive weight mechanism is adopted to update the particle velocity, and then an energy-aware local fitness first mechanism is designed to update the particle location in order to improve the efficiency. The simulation results indicate that the proposed algorithm significantly improves the resource utilization and reduces the energy consumption cost of virtualized data center, while meeting the application performance constraint and application constraints.2. To resolve the dynamic resource management and provisioning multi-objective collaborative optimization problem of IaaS cloud services hosted in large-scale virtualized data centers, a virtualized data center-oriented dynamic resource management mechanism is presented, and the core two-stage resource reconfiguration multi-objective provisioning optimization algorithm is implemented. The proposed algorithm takes account of the total energy consumption, resource utilization, and system reliability while subject to application constraint and performance constraint. Above all, it presents a novel hybrid multi-objective tabu search algorithm to obtain the Pareto optimal solution set, and then figures out the optimal solution to resource reconfiguration optimization problem using the TOPSIS method of multi-attribution decision theory. The comparison of simulation experiments show that the proposed algorithm achieves multi-objective dynamic resource provisioning optimization of virtualized data center resources, while meeting the performance constraint and application constraint of cloud services, which provides a useful reference for dynamic resource management and provisioning optimization of virtualized data center.3. Aiming to resolve the confliction between the Service Level Agreement (SLA) constraints of SaaS cloud service and resource provisioning cost, a SaaS cloud service request model with SLA constrains is established to quantify the performance constraint, and then a cost-aware cloud service request scheduling algorithm based on dynamic reuse is presented. According to personalized characters of cloud service requests, different virtual machine instance configuration and their pricing models, combined with the current system load, our proposed algorithm can not only rent and reuse diverse heterogeneous virtual machine instances on demand to achieve optimal scheduling of dynamic cloud service requests in reasonable time, but also can minimize the rental cost of the overall infrastructure for increasing cloud service providers’ profits while meeting SLA constraints. The simulation results verify that the proposed scheduling algorithm significantly reduces the resources lease cost, improves the resource utilization and operation profit of SaaS providers.