Deep Reinforcement Learning solution for Scheduling critical notifications in a Digital Twin cluster

Abstract

A Scheduling approach for a Critical Monitoring System in a Digital Twin (DT) cluster based on Deep Reinforcement Learning (DRL) is presented. Recent advances in the DRL field inspired us to research how to build a solution that learns to manage the cloud container’s resources directly from experience. The paper presents a multi-objective Scheduling approach for containerized microservice Critical Notification system applications based on DRL (SCN-DRL), where Neural Networks (NN) are used RL Agents. This paper implements, compares and evaluates three Neural Networks (NN) that: a) provide scheduling of the DT cluster’s notification jobs, b) outperform state-of-art heuristics, and c) keep steady performance when notification workload increases from 10 to 90%. Furthermore, resilience to resource container failures is a critical component of the distributed system; our proposed research shows that SCN-DRL is resilient to sudden resource drops by 10%.

Cite as: M. Vrbaski, M. Bolic, S. Majumdar, Deep Reinforcement Learning solution for Scheduling critical notifications in a Digital Twin cluster, 2023.
Download Paper