A State-of-the-Art Review of Intelligent, Dependency-Aware Autoscaling for Microservice Architectures

Abstract

Managing resource allocation for microservice architectures is a significant challenge, as traditional autoscaling methods often lead to costly overprovisioning or performance degradation that violates Service Level Objectives (SLOs). This thesis provides a comprehensive stateof- the-art review of smart autoscaling strategies designed to overcome these limitations. The core of this work is a systematic survey and analysis of cutting-edge research that leverages advanced Artificial Intelligence (AI), with a particular focus on Reinforcement Learning (RL) and Graph Neural Network (GNN) based approaches. By examining and comparing prominent academic and industry solutions, this review creates a structured map of the current research landscape. This comparative analysis serves as a valuable resource for researchers and practitioners aiming to understand, implement, or advance the field of intelligent resource management for modern cloud-native applications.