Energy-aware scientific workflow scheduling in cloud environment

Cloud computing represents a significant shift in computer capability acquisition from the former ownership model to the current subscription approach. In cloud computing, services are provisioned and released in a distributed environment and encourage researchers to further investigate the benefits of cloud resources for executing scientific applications such as workflows. Workflow is composed by a number of fine-grained and coarse-grained tasks. The runtime of fine-grained tasks may be shorter than the duration of system overheads. These overheads can be reduced by merging the multiple fine-grained tasks into a single job which is called task clustering. Clustering of the task is itself a big challenge because workflow tasks are dependent on each other either by data or control dependency. Further, workflow scheduling is also critical issues which aimed to successfully complete the execution of workflow without compromising the agreed Quality of Service parameters such as deadline, cost, etc. Energy efficiency is another challenging issues and energy-aware scheduling is a promising way to achieve the energy-efficient cloud environment. Traditional research in workflow scheduling mainly focuses on the optimization constrained by time or cost without paying attention to provide complete framework for workflow scheduling. The main contribution of this study is to propose a novel scheduling framework that provide a step by step solution for workflow execution while considering the mentioned issues. In order to minimize energy consumption and total execution cost, power-aware dynamic scheduling algorithms are designed and developed that try to execute scientific applications within the user-defined deadline. We implement the task clustering and partial critical path algorithm which helps to forms the jobs of fine-grained tasks and recursively assign the sub-deadlines to the task which are on the partial critical path. Further, to improve the energy efficiency, we implement Dynamic Voltage and Frequency Scaling (DVFS) technique on computing nodes to dynamically adjust voltage and frequency of the processor. Simulation is performed on Montage, CyberShake, SIPHT, LIGO Inspiral Analysis scientific applications and it is observed that the proposed framework deal with the mentioned issues. From the analysis of results it is observed that using clustering and DVFS technique transmission cost and energy consumption is reduced at considerable level.