|Institution:||University of Technology, Sydney|
|Full text PDF:||http://hdl.handle.net/10453/34477|
Elasticity and cost-effectiveness are two key features for ensuring that cloud-based web services appeal to more businesses. However, true elasticity and cost-effectiveness in the pay-per-use cloud business model has not yet been fully achieved. The explosion of cloud-based web services brings new challenges to enable the automatic scaling up and down of service provision when the workload is time-varying. This research studies the problems associated with these challenges. It proposes a novel scheme to achieve optimised auto-scaling for cloud-based web services from three levels of cloud structure: Software as a Service (SaaS), Platform as a Service (PaaS), and Infrastructure as a Service (IaaS). At the various levels, auto-scaling for cloud-based web services has different problems and requires different solutions. At the SaaS level, this study investigates how to design and develop scalable web services, especially for time-consuming applications. To achieve the greatest efficiency, the optimisation of service provision problem is studied by providing the minimum functionality and fastest scalability performance concerning the speed-up curve and QoS (Quality of Service) of the SLA (Service-Level Agreement). At the PaaS level, this work studies how to support dynamic re-configuration when workloads change and the effective deployment of various kinds of web services to the cloud. To achieve optimised auto-scaling of this deployment, a platform is designed to deploy all web services automatically with the minimal number of cloud resources by satisfying the QoS of SLAs. At the IaaS level for two infrastructure resources of virtual machine (VM) and virtual network (VN), this research focuses on studying two types of cloud-based web service: computation-intensive and bandwidth-intensive. To address the optimised auto-scaling problem for computation-intensive cloud-based web service, data-driven VM auto-scaling approaches are proposed to handle the workload in both stable and dynamic environments. To address the optimised auto-scaling problem for bandwidth-intensive cloud-based web service, this study proposes a novel approach to predict the volume of requests and dynamically adjust the software defined network (SDN)-based network configuration in the cloud to auto-scale the service with minimal cost. This research proposes comprehensive and profound perspectives to solve the auto-scaling optimisation problems for cloud-based web services. The proposed approaches not only enable cloud-based web services to minimise resource consumption while auto-scaling service provision to achieve satisfying performance, but also save energy consumption for the global realisation of green computing. The performance of the proposed approaches has been evaluated on a public platform (e.g. Amazon EC2) with the real dataset workload of web services. The experiment results demonstrate that the proposed approaches are practicable and achieve superior performance to other benchmark methods.