Reducing energy footprint in cloud computing: a study on the impact of clustering techniques and scheduling algorithms for scientific workflows

The concept of scientific workflow makes it possible to link and control different tasks to carry out a complex treatment. The complicated workflow is generated by scientific distributed applications that may contain thousands of tasks. This high number of tasks requires important computation capabilities over the cloud datacenters. The rate of tasks that require execution by the cloud may lead to hosts’ overloading, which may increase the energy consumption and makespan of workflows. As a result, efficient techniques are necessary to save energy and time. Task clustering is an efficient technique that involves combining multiple tasks into one unit, called a job, to reduce the resource allocation time for the workflow’s tasks and consequently reduce the makespan. On the other hand, the scheduling of tasks’ execution in cloud hosts may have an impact on energy consumption and makespan, so it is asked to wisely integrate the scheduling algorithms into the computation of workflows. In this study, we analyze the contribution of task clustering techniques and scheduling algorithms on energy consumption and makespan during the computation of scientific workflows by the cloud’s infrastructure. For this purpose, we used WorkflowSim, an open-source cloud simulator providing workflow level support, task scheduling, and clustering techniques. The simulations’ results conclude that clustering techniques affect the energy consumption and Makespan regardless of the deployed scheduling scheme, however some combination of both the scheduling and clustering techniques may reduce the Makespan and consequently reducing energy consumption; their impact is more related to the nature of the running scientific workflow in the cloud. The main simulations’ results observation shows that Vertical clustering and MaxMin algorithms are more suitable for saving energy.