عليرضا افرا

Serverless computing has garnered significant attention from cloud service providers and software developers in recent years. This model, also known as Function as a Service (FaaS), enables developers to run their functions in cloud environments without the need to manage complex infrastructure. In this model, the management of computing resources is entirely handled by the service provider. On the other hand, cloud providers are continuously seeking to optimize their resource utilization to reduce costs and increase efficiency. However, proper scheduling of functions to computing resources remains a significant challenge in this domain. Most existing scheduling algorithms primarily focus on load balancing across resources and typically rely on simple, default techniques. These algorithms are often incapable of addressing the dynamic needs of serverless environments, which can lead to reduced overall system performance and suboptimal function execution. Moreover, many of these methods do not pay sufficient attention to the specific requirements of functions, such as the necessary software packages. For example, the absence of required packages in computing containers can lead to delays in function execution, ultimately increasing response time. In this research, we propose a package-aware scheduling approach based on deep reinforcement learning (DRL) in edge serverless computing environments. In this method, the scheduler automatically assigns functions to computing containers that already have the required software packages installed. This reduces the time spent on package installation, improves system efficiency, and decreases function response time. Our proposed method employs a deep neural network to enhance the scheduler's behavior over time, using past experiences to make better decisions. To eva‎luate the efficiency of the proposed method, we conducted experiments using a simulator based on the OpenWhisk platform, simulating conditions similar to operational environments. The results demonstrate that our proposed method achieves improvements in reducing response time and lowering resource consumption costs compared to existing approaches. These findings highlight the significant potential of reinforcement learning methods in enhancing function scheduling in serverless environments.

محاسبات ابري , محاسبات بدون سرويس دهنده , زمانبندي توابع , يادگيري تقويتي عميق.

Cloud Computing , Serverless Computing , Function Scheduling , Deep Reinforcement Learning

Alireza Afra

Mehrdad Ashtiani