Research On Scheduling Techniques To Minimize Operational Costs In The Cloud

In recent years, the cloud computing technology, with its virtually infinite resources and pay-per-use cost model, is being adopted in the realm of business, engineering, and healthcare. Depending on the relationship between resource owners and users, cloud computing can be classified into three modes:public cloud, private cloud, and hybrid cloud. As the major mode of cloud computing, public cloud has three players: business users, service providers, and cloud providers. These players are often come from different parties with their own interests. Specially, the goal of business users and service providers is to minimize resource rental costs without violating service level agreements. On the other hand, cloud providers seek to achieve high resource utilization and reduce energy consumption while meeting resource requirement.Motivated by the above requirements, we focus on developing scheduling algorithms for business users, service providers, and cloud providers, respectively. Correspondently, we solve three different problems in public cloud environments:the first is how to minimize resource rental costs for running parallel tasks; the second is how to minimize resource rental costs for running interactive services and batch jobs; the third is how to reduce energy consumption for consolidating applications. The main contributions of this thesis are as follows:(1) To schedule parallel tasks for business users with minimum resource rental costs, we give a formulation for the problem of parallel task scheduling in the cloud. Then, we propose a new cost-conscious parallel task scheduling algorithm. It first maps tasks to the most cost-efficient instances based on the theory of Pareto Dominance, and then reduces the resource rental costs for non-critical tasks. The experimental results show that our algorithm can substantially reduce resource monetary costs while producing execution time as good as the best known task-scheduling algorithm can provide.(2) To schedule interactive services and batch jobs for service providers with minimum resource rental costs, we give a new formulation for scheduling interactive services and batch jobs. Then, we develop a profit-driven scheduling algorithm to minimize the resource rental costs while still meeting any performance-based service level agreements. The experimental results demonstrate that our scheduling algorithm significantly outperforms the existing scheduling approaches for just interactive services or just batch jobs.(3) To reduce power consumption for cloud providers. We use an Online Coloring Bin Packing problem to model the consolidation problem and devise an effective application-aware approximation algorithm to find a near-optimal solution. A formal analysis also demonstrates the approximation ratio of this algorithm. The experimental results show that the proposed algorithm provides significant savings of energy compared to several benchmark algorithms.