A Proactive Cloud Application Auto-Scaler using Reinforcement Learning

This work explores the use of reinforcement learning to design a proactive cloud resource auto-scaler that is able to predict usage across a distributed microservice application. The focus is on serving time-sensitive workloads, e.g., industrial automation, connected XRNR (eXtended Reality/Virtual Reality), etc., where each job has a deadline and there is some cost associated with missing a deadline. A simple workload model, as well as a microservice application model, is presented. A reinforcement learning agent is trained to identify workloads and predict needed utilization for identified service chains. The results are compared to standard purely reactive techniques.