A Genetic-Algorithm-Based Framework for Green Task Scheduling and Energy Optimization in Data Centers

Abstract

Data centers already draw a fast-rising slice of global electricity, and regulators and export markets are watching their carbon footprint more closely every year. This paper presents a genetic-algorithm (GA) framework for green task scheduling that jointly minimizes energy consumption, carbon emissions, and service-level-agreement (SLA) violations across a geo-distributed data-center environment with time-varying grid carbon intensity. What sets it apart from prior carbon-aware schedulers is a dedicated carbon-data governance layer sitting underneath the optimizer: a unified data catalog, built on multi-source interoperability, that feeds the scheduler carbon-intensity signals whose source, usage, and processing history can all be checked after the fact. Each batch or deferrable task is encoded as a mapping to a server and a time slot, and the GA evolves populations of these mappings under a normalized weighted-sum fitness function using tournament selection, two-point crossover, adaptive mutation, and elitism. We test the framework in a discrete-event simulation with up to two thousand tasks and two hundred heterogeneous servers, driven by a diurnal carbon-intensity curve and Google-cluster-style workloads. Against Round-Robin, First-Fit and Best-Fit greedy heuristics, particle swarm optimization, and a carbon-agnostic GA, the proposed carbon-aware GA cuts carbon emissions by roughly eighteen to twenty-eight percent and energy consumption by twelve to twenty percent, with SLA impact staying close to negligible. Ablation and sensitivity studies point to the same conclusion: governed carbon data and careful weight tuning, together, are what make the low-carbon scheduling outcomes both strong and defensible.