Short-Term Scientific Mission

Black-Box Multi-Objective Problem Modeling and Benchmarking for Cloud Resource Management

Main theme: Single- and Multiobjective Optimisation
Grantee: Faraz Shaikh, University of Perugia, Italy
Host: Filip De Turck, Ghent University, Belgium

Start date: 2026-08-03
End date: 2026-08-31
Awarded: 2026-07-09

Description

Autoscaling containerized services in edge-cloud environments requires balancing service latency, infrastructure cost, and energy consumption simultaneously, making it a natural multi-objective black-box optimization problem over a mixed discrete-continuous decision space. Prior work has produced a Pareto-optimal autoscaling framework based on multi-objective reinforcement learning, yet no comparison with classical randomized optimization algorithms has been conducted. The goal of this STSM is to formalize this problem as a ROAR-NET benchmark instance and compare the reinforcement learning-based Pareto approximation against NSGA-II and MOEA/D. Activities will include defining the evaluation protocol, running comparative experiments using anytime hypervolume curves, characterizing oracle noise, and submitting the benchmark specification to the ROAR-NET library.