Virtual Infrastructure Planning For A Workflow With Multiple Overlapping Deadline Constraints In Cloud

Cloud providers offer various types of virtual infrastructures (e.g. Virtual Machines, Dockers) to consumers in a pay-as-you go manner. For an application represented by a complex workflow, it is difficult for the consumer to decide what type of infrastructure they need to meet the Quality-of-Service (QoS) requirements and achieve objectives like monetary cost optimisation. We call such problem as the virtual infrastructure planning problem in cloud. Most existing studies focus on only one single deadline-constrained workflow planning. The single deadline constraint usually refers to a global deadline since the start execution of a workflow to its finish. However, such simple model cannot allow a time critical application to specify internal deadlines inside the workflow. To allow such flexibility for consumers, we propose a multi-deadline model and consider the virtual infrastructure planning problem for workflow with multi-deadline constraints. To solve the problem efficiently, we propose a Multiple-Deadline overlapping-based Infrastructure Planning (MDIP) algorithm. In MDIP, we define the criticality of tasks by analysing the deadline coverages. Then we rank the tasks in the workflow by their criticality and assign them with VM type of better performance. To evaluate the effectiveness of our algorithm, we compare it with a meta-heuristic solution (Genetic Algorithm-based) and a hybrid solution combining critical path and genetic algorithm. Simulated experiments show our approach can achieve better results than existing solutions and more efficient than GA-based solution.