An Advantage Actor-Critic and Proximal Policy Optimization Model for Adaptive Scheduling of Heterogeneous Cloud Workloads

Abstract

Efficiently scheduling heterogeneous cloud workloads in modern computing environments is a formidable challenge. Managing diverse resource needs, intricate task dependencies, and optimizing resource allocation are key concerns in this context. To address these challenges, this paper introduces a novel approach that combines the Advantage Actor-Critic (AAC) and Proximal Policy Optimization (PPO) models. AAC accurately estimates task dependencies, while PPO optimizes task scheduling operations. The proposed model leverages essential parameters such as CPU utilization, memory requirements, storage demands, network bandwidth, task deadlines, and task dependencies. The results of experiments conducted demonstrate the efficacy of the proposed approach. It achieves a 4.9% reduction in makespan, indicating faster task completion, an 8.5% increase in resource utilization, showcasing improved resource efficiency, a 2.5% reduction in energy requirements, extending the cloud deployment lifespan, and a 4.9% increase in the deadline hit ratio, highlighting the model's proficiency in meeting task deadlines. In conclusion, the novel scheduling approach, integrating AAC and PPO models, offers substantial advantages over existing methods. Precise task dependency estimation and optimized scheduling lead to enhanced performance metrics, including reduced makespan, augmented resource utilization, increased throughput, and improved deadline adherence. These findings have promising implications for real-world applications, underscoring the solution's potential impact on contemporary computing environments.