Hybrid Cloud Resource Provisioning For Real-time Workflows Scheduling

The key to the problem of workflow scheduling in cloud is to match tasks and various types of resources.The cloud provider supplies abundant resource types and rental models.A hybrid resource provisioning method with on-demand instance and spot instance is considered in this thesis.On one hand,on-demand instance is hourly priced and the performance is stable,but it is relatively expensive;on the other hand,spot instance is divided into the general type and spot block.The general type is inexpensive and the price fluctuates over time.It will be terminated due to out-of-bid failure.Spot block offers a variety of rental intervals,with constant prices and stable performance.We study the hybrid resource provisioning for real-time workflows scheduling problem.Firstly,two kinds of hybrid resource provisioning models are put forward,and then the math-ematical models are established respectively.To address the problem,two new heuristics are proposed.Then the main components of the algorithm are described:task scheduling,dynamic resource provisioning and spot monitoring.The task scheduling is divided into four parts name-ly ready task sequencing,slack time calculation,idle-time slots searching and rules-based task scheduling.Four rules for task sequencing are developed according to the workflow priority and topology.By using the slack time calculation and idle-time slots searching,the resource utilization is improved.Different scheduling rules are raised in the rules-based task scheduling.Considering interval-priced resources,two releasing strategies are designed.Spot monitoring is responsible for monitoring the request waiting time and out-of-bid failure.Parameters and related components of the proposed algorithm are verified using the ANO-VA technique.Through a large number of standard scientific workflow instances,the experi-mental results show that our proposed algorithm outperforms the comparison algorithms under different deadline levels.