Stochastic Optimization Models for Energy-Aware Cloud Resource Allocation Using Simulation-Derived Scenarios

Abstract

Abstract: Cloud resource allocation decisions are typically made under significant uncertainty about future workloads, yet many operations research formulations treat demand deterministically or rely on simplistic safety margins. This study develops deterministic and stochastic optimisation models for energy-aware virtual machine placement, server activation, and scheduling in cloud data centres. Scenario sets for future demand are derived from the stochastic modelling and scenario generation framework, ensuring realistic temporal and cross-sectional variability. The deterministic and two-stage stochastic models are implemented as linear and mixed-integer programs in MATLAB, with objective functions that combine energy consumption, operating cost, and penalties for service-level agreement (SLA) violations. Numerical experiments are calibrated using CloudSim simulations to approximate realistic load and power profiles. The results show that two-stage stochastic models significantly reduce expected energy and SLA penalties compared with deterministic baselines, particularly under highly variable and bursty workloads. Sensitivity analyses explore the effects of scenario set size, risk measures, and penalty parameters. The study demonstrates the practical value of scenario-based stochastic programming for cloud resource allocation and provides a foundation for integrating such models into broader hierarchical control architectures. Keywords: stochastic optimization, cloud computing, resource allocation, energy efficiency, mixed-integer programming, scenario-based planning, service-level agreements, stochastic programming