Toward optimizing scientific workflow using multi-objective optimization in a cloud environment

Abstract

AbstractScientific workflows are a common and critical part of scientific computing, involving complex computations and oversized and distributed computing resources. Efficient workflow execution requires scheduling algorithms considering task dependencies, resource requirements, and deadlines. Cloud computing provides an innovative architecture for extensive heterogeneous computing services. However, scheduling hybrid cloud resources with deadline restrictions while observing QoS standards is an NP-complete task. Mapping workflow tasks to virtual machines and determining the optimal schedule order is a challenging aspect of cloud computing. By executing task requests on the most advantageous virtual machine in the resource pool, energy consumption, overall execution time, and computing costs can be reduced. This research aims to identify the best location to process applications using user’s demand and priority. A multi-objective genetic algorithm is proposed to achieve this objective, which considers conflicting objectives such as time, energy, cost, and deadline. The algorithm initializes an efficient ranking heuristic approach and predicts the earliest finish time (PEFT) using the Bayesian approach to improve the Pareto fronts. This approach enhances the VM migration of cloud-based tasks and optimizes the search space for conflicting objectives. Experimental findings show that the proposed approach reduces cost by 5–6% and time delay by 8% compared to existing approaches. The proposed approach offers an effective solution for scheduling scientific workflows on cloud computing resources while considering various QoS standards. The results demonstrate the potential of multi-objective genetic algorithms for optimizing workflow scheduling in cloud computing environments.

Keywords

• workflow scheduling
• load balancing
• Artificial Intelligence
• optimization
• ranking
• ChatGPT