Multi-Objective Cloud Resource Optimization with NSGA-III: Design, Deployment, and Open Problems

Abstract

Cloud computing has become the backbone of modern digital infrastructure, yet the simultaneous optimization of competing operational objectives, makespan, monetary cost, and energy consumption remains an open and practically significant research challenge. Existing commercial schedulers and the majority of academic proposals address only one or two of these objectives, discarding trade-off information that is essential for heterogeneous operator preferences. This paper presents a conceptual framework for applying the Non-dominated Sorting Genetic Algorithm III (NSGA-III) to the multi-objective cloud resource scheduling problem, with the explicit goal of informing the design of a publicly deployable optimization service. Drawing on established theoretical properties of NSGA-III, published workload characterization studies, and cloud service architecture literature, we identify the core design decisions that a realization of such a framework must resolve: chromosome encoding for workflow-structured workloads, cloud-topology-aware genetic operators, scalability management under real-time SLA constraints, and Pareto-front visualization for non-expert decision-makers. We further map four concrete implementation pathways, open-source library, cloud-provider plugin, SaaS scheduling API, and embedded data-center advisors and analyze each against criteria of deplorability, latency, governance, and reproducibility. It synthesizes existing knowledge to establish the conceptual and architectural foundations upon which future empirical work can be built and articulates a research agenda of seven open problems that must be solved before NSGA-III-based multi-objective cloud scheduling can reach production at scale.