Problem
Microservices make it possible to scale individual components independently, but their internal dependencies make it difficult to predict global performance and make real-time elasticity decisions.
Approach
The proposal combines LQN models, which capture interactions among services, with GNNs capable of learning representations over application graphs.
Expected outcome
A performance predictor and an elasticity controller to dynamically adapt resources while meeting quality-of-service requirements.
Summary
Cloud computing and edge computing have transformed how services and applications are executed. In this context, microservices-based architectures enable modular, reusable and scalable systems, while also introducing new challenges for efficient resource management.
The project addresses elasticity in microservices-based applications: the ability to dynamically, autonomously and optimally adapt allocated resources under variable workloads. To this end, it proposes an elasticity controller that integrates a performance predictor based on Graph Neural Networks and Layered Queueing Networks.