Dynamic Workflow Scheduling in Cloud using Neural Network based Multi-Objective Evolutionary Algorithm

Abstract

Workflow is a series of jobs that are executed in order to complete a specific activity where the jobs are often dependent on each other. Data transfer that might occur between such jobs results into the creation of a workflow that aims at utilising resources for workflow tasks by optimising the use of cloud resources. Few of the existing single objectives workflows scheduling solutions have linearly combine multiple objectives to get a multi-objective solution, but it might not be able to model the real-world problem efficiently for certain conditions where the environment is dynamic in nature. Hence, a Neural Network based Multi-Objective Evolutionary Algorithm (NN-MOHEFT) that solves the multi-objective workflow scheduling issues in a dynamic environment was proposed in this article. The NN-MOHEFT learns the pattern behind changing Pareto optimal front for successive environment and tries to predict the Pareto optimal front for the next environment from the input Pareto optimal set of the current environment. The proposed NN-MOHEFT algorithm is at par with the original constructs when it comes to the hypervolume of objectives generated. It generates 10% more non-dominated solutions as compared to the original construct.