Optimal Resource Provisioning and Efficient Workflow Scheduling Algorithms in IaaS Cloud

Abstract

The scientific applications have become ever more compute-intensive, elevating newlinethe need for storage and large-scale computational resources, either in the cluster of newlineHigh-Performance Computing (HPC) platforms or in the cloud. Of late, the cloud has newlinebecome a viable environment for hosting large-scale business and scientific workflow newlineapplications due to cloud dynamics such as heterogeneous resource types, elasticity, ondemand newlineprovisioning, pay-per-use cost model, and virtualization. It offers a low capital newlineas well as a low barrier alternative to operate dedicated infrastructure. Evidently, commercial newlineclouds are now enabling universal access to their services, which are previously newlineavailable only to large well-funded research groups. Even though the potential benefits newlineof cloud are evident, there are still some associated challenges when trading off the newlineoptimal execution efficiency and costs. newlineComplex scientific applications are represented naturally in the form of workflows newlinefor scheduling and run time provisioning. The adoption of workflow in the scientific newlinecommunity has led to the acceleration of scientific discovery. Such workflow scheduling newlinewith user-specified Quality of Service (QoS) constraints is becoming an even more newlinechallenging issue due to the dynamic and uncertain nature of the cloud. The mapping of newlineworkflow tasks onto a set of provisioned resources is an example of the general scheduling newlineproblem and NP-complete. Due to their large-scale nature, scheduling algorithms newlineare essential for efficient automation of their execution in the cloud, thereby facilitating newlineand accelerating the pace of scientific advancement. In addition, certain run time constraints newlinelike the cost of the computation, i.e., budget and the time of the computation to newlinecomplete, i.e., deadline must be met. A well-managed budget and deadline constraints newlinescheduling are required to improve the system s performance and end-user satisfaction. newlineThis thesis investigates the scientific workflow scheduling problem in the cloud newlineby addressing the fundamenta