Research On Cloud Task Scheduling Optimization Based On Deep Reinforcement Learning

Cloud computing has become a new trend in the use of computing resources.It allows users to purchase on demand and remotely use various computing resources in a configurable resource sharing pool,reducing computing costs and improving user experience.The quality of task scheduling in the cloud computing environment not only affects the efficiency of the system,but also affects the user experience.In order to obtain a near-optimal solution within an acceptable time,a large number of task scheduling methods have been proposed in the past decades.However,almost all methods are designed to handle batch workloads due to the computational overhead in the optimization process.In contrast,real-time task scheduling is rarely studied.While some traditional methods can be used to make real-time decisions in cloud computing environments,these methods are inefficient when dealing with complex transactional workloads.With the development of machine learning,deep learning has become an effective method to solve complex scheduling problems in different fields.Deep reinforcement learning can optimize schemes without any prior knowledge of the system,and in recent years,it has shown remarkable ability in dealing with scheduling problems in cloud computing.On this basis,this paper leverages deep reinforcement learning technology to further explore the task scheduling problem in the cloud computing environment,aiming to propose a strategy and model with better scheduling effect.The specific research content is as follows:1.A Deep Reinforcement Learning-based Preemptive Approach for Cost-aware Cloud Job SchedulingNone of the known deep reinforcement learning scheduling frameworks dealing with real-time cloud tasks in recent years considers additional optimization opportunities for assigned tasks in their framework.In view of this situation,this paper will introduce a new preemptive method based on deep reinforcement learning.Specifically,it improves the training effect of existing strategies by adding an effective task preemption mechanism,and on this basis,achieves the training goals of reducing task execution costs and meeting the expected response time of users as much as possible.The experimental results show that,compared with the existing cloud scheduling model,this method can perform scheduling tasks better under different real-time workloads,and is better than existing scheduling algorithms.2.Deep Adversarial Imitation Reinforcement Learning for QoS-aware Cloud Job SchedulingConsidering that the trajectories of tasks in the cloud are always long,current deep reinforcement learning-based solutions face challenges in finding high-reward trajectories,and thus suffer from issues such as suboptimal scheduling strategies.In order to improve this problem,a deep adversarial imitation reinforcement learning framework will be proposed in the subject.The framework seamlessly combines adversarial imitation learning and deep reinforcement learning,focusing on scheduling user requests in a way that maximizes task success while significantly reducing task response time.Compared with the latest deep reinforcement learning-based methods,this method can serve as an expert to cache high-return work trajectories,and provide instant guidance for deep reinforcement learning agents to learn better policies during training,thus making it more efficient,especially for cloud tasks whose trajectories are always long.Experimental results show that under different real-time workloads and computing resource configurations,this method can generally outperform existing cloud task scheduling methods with the goal of maximizing the success rate of timesensitive tasks and minimizing the average response time.