Reinforcement learning microservices scheduler in intelligent edge computing

Abstract

The proliferation of internet of things (IoT) devices and resource-intensive applications has necessitated the development of intelligent edge computing frameworks. These frameworks aim to address challenges in the resource management, service latency, and data privacy of IoT devices. This research investigates the complex problem of microservice scheduling within intelligent edge computing environments. The focus is on optimising quality of service (QoS) metrics such as the latency, network bandwidth utilisation, and energy consumption during execution of resource-intensive applications.

To address this challenge, a novel approach called the Bi-generic A2C Microservice Proxy Policy (BAMPP) is proposed. It leverages reinforcement learning (RL) principles to optimize microservice deployment in dynamic Edge-Cloud ecosystems. BAMPP uniquely considers the intricate inter-dependencies among microservices and adapts to user mobility in real-world scenarios.

This research utilises a simulation platform to reproduce the intelligent edge computing environment, integrating real-world datasets to evaluate the performance of BAMPP against comparative algorithms. The research focuses on three key research points: identifying crucial factors influencing microservice scheduler performance, leveraging RL for optimised scheduling, and assessing the impact of random user mobility on service deployment.

The results demonstrate BAMPP's superior performance in reducing energy consumption, minimizing network usage, decreasing execution and migration latency, and enhancing reliability in microservice scheduling compared to current systems. This research contributes to the field of intelligent edge computing by introducing a novel modeling approach, developing an advanced algorithm for joint optimization of scheduling and resource management, and providing comprehensive performance evaluations using realistic simulations.

The results of this study have important ramifications for raising the effectiveness and performance of microservice applications in intelligent edge environments, potentially leading to cost savings, enhanced sustainability, and widespread implementation across diverse edge computing scenarios.