Auto-scaling for deadline constrained scientific workflows in cloud environment

Cloud computing is an ideal platform for scientists to realize large-scale deadline constrained Scientific Workflows (SWf), since it often require hours to complete its execution. Efficient resource provisioning and task scheduling of SWf play a vital role in cloud computing. Moreover, mapping of computing resources and precedence constrained task's of SWf to meet user specified deadline with minimum execution cost is crucial. Therefore, an efficient auto-scaling mechanism is extremely essential. In this paper, a task scheduling strategy, auto-scaling architecture and auto-scaling method for SWf are proposed, that guarantees the execution of different SWf within deadline in a cloud environment. Slot time and idle-time of the VM, and sub-deadline of the SWf tasks is considered to achieve the objective. Experiments include two well known SWf, namely, Epigenome and Cybershake of different dependencies and data file sizes. We evaluate our approach by analysing failure of tasks at different intervals of deadline, makespan and total cost incurred to execute SWf of different workflow sizes. Simulation results proved that the proposed auto-scaling method not only reduces total number of failure tasks, but also minimizes makespan and total execution cost of SWf.