Performance evaluation of different scheduling approaches in real-time heterogeneous serverless edge environment

Serverless edge computing is quickly becoming a key player in improving distributed systems, especially for the Internet of Things (IoT). This paper looks into how to schedule tasks in these systems, where it’s crucial to process data close to where it’s created. The study carefully examines a wide range of methods for task scheduling, from foundational algorithms like Random and Round Robin to advanced techniques including Reinforcement Learning (RL) and Consistent Hashing (CH). Through a custom-built simulation environment, leveraging Docker, the research simulates real-world conditions to measure the performance of these scheduling methods. Key metrics such as cold-start frequency, response time and failure rates are analyzed to ascertain the most effective strategies for diverse operational scenarios. The main goal is to fill the empirical research gap, providing actionable insights into the real-world applicability of these scheduling methods in serverless edge computing. The results highlighting that Reinforcement Learning (RL) methods exhibit promising adaptability and performance, particularly in dynamic and resource-constrained environments. Advanced techniques like CH-BL outperform others by near 10% in efficiency, especially in response time, while traditional algorithms face challenges with cold-starts and optimal resource utilization.

Performance evaluation of different scheduling approaches in real-time heterogeneous serverless edge environment Keywords:

serverless edge computing , deep reinforcement learning , scheduling , consistent hashing