Task Allocation in the Edge Cloud Scenario Incorporating Customized Requirements

As Edge-Cloud architectures continue to evolve, users increasingly demand customized task scheduling solutions. Traditional task scheduling aims at optimizing resource utilization and task prioritization. While conventional methods use random task arrival and first-come-first-served (FCFS) allocation or machine learning techniques like Reinforcement Learning (RL), they often lack consideration for individual task preferences, such as cost-time trade-offs, wait time, and user experience. In our paper, we introduce a framework for task scheduling that takes into account user-specific non-essential preferences, such as cost versus execution time and waiting time. This framework allows for parameter customization to meet these specific requirements. We also compare results from RL-based approaches with our framework, examining the impact of using actual task arrival data versus random data, and the performance of non-essential parameters. Our prototype in an edge-cloud environment, using a doctor-patient scenario, demonstrates that our simulation-based technique offers flexibility in parameter selection, resulting in improvements in fairness, reduced task waiting times, better load balancing, and minimized server idle time. Furthermore, our framework shows sensitivity to changes in input datasets, which is a limitation of the RL approach.