چكيده لاتين
In todayʹs era, cloud computing has emerged as a new approach, offering on-demand computing, storage, and networking services, bringing flexibility, scalability, and efficiency to businesses and organizations. However, the optimal management of computing resources in cloud environments, given the increasing complexity and workload, has become a major challenge. Task scheduling, as one of the most important aspects of resource management, plays a key role in ensuring efficiency, reducing costs, and increasing user satisfaction. In this problem, various tasks with different computational demands and deadlines enter the system and must be scheduled in a way that minimizes the costs associated with energy consumption and delays in task processing. This challenge, due to the dynamic and random nature of task arrivals and the limitations of computing resources, requires the development of intelligent and efficient scheduling algorithms. In this research, an adaptive scheduling algorithm called AEDLDF-PPO is proposed to solve the task scheduling problem in cloud computing. By leveraging the strengths of deep reinforcement learning and prioritization rules, this algorithm can learn optimal scheduling policies under various traffic conditions and minimize system costs. To this end, the scheduling problem is first formulated as a dynamic program with uncertain transition probabilities, and a comprehensive mathematical model for the cloud computing system is presented. Then, considering the limitations of traditional scheduling methods and existing deep reinforcement learning algorithms, the proposed AEDLDF-PPO algorithm is introduced. This algorithm consists of two main stages: the traffic classification module and the adaptive task scheduling module. The traffic classification module uses a network to classify the current system traffic level as light or heavy, and then the adaptive task scheduling module, which includes two EDLDF-PPO scheduling policies for heavy traffic and one EDF policy for light traffic, selects the appropriate policy for task scheduling at each stage based on the output of the traffic classification module. The performance of the proposed AEDLDF-PPO algorithm has been evaluated using extensive simulations in cloud computing environments with various traffic conditions. The simulation results show that this algorithm outperforms the LLF-PPO, EDF-PPO, and RR-PPO scheduling algorithms in all the scenarios considered. By intelligently adapting between the EDLDF-PPO and EDF policies based on traffic conditions, this algorithm can minimize overall system costs. In general, the results of this research show that the proposed AEDLDF-PPO algorithm is an efficient and effective approach for task scheduling in cloud computing environments. This can help improve efficiency and reduce costs by up to 14% compared to the baseline algorithm in cloud computing systems, and consequently, contribute significantly to the development and expansion of these technologies.
