Intelligent deep reinforcement learning-based real-time job schedulerfor energy-efficient cloud data centres in Botswana

Cloud computing’s growth has increased data-centre energy demand, a pressing issue in supply-constrained, coal-reliant grids such as Botswana’s. This study proposes and evaluates an Intelligent Deep-Reinforcement-Learning–Based Scheduler (IDBS) that makes real-time job-to-VM placement decisions to minimise energy use while maintaining quality of service (QoS). We formulate scheduling as a Markov decision process with explicit state/action spaces for heterogeneous VMs and mixed job types and train a Deep Q-Network with experience replay and a target network for stability. In simulation with 8,000 Poisson-arriving jobs spanning I/O- and compute-intensive workloads on heterogeneous VMs, IDBS is benchmarked against Random, Round Robin, Earliest, and JingDQN. Across scenarios that vary arrival rates, job-type proportions, and VM-type proportions, IDBS consistently lowered energy cost, reduced average response time, raised job-completion success, increased VM utilisation, and sustained higher QoS relative to baselines. These findings indicate that DRL-based, QoS-constrained scheduling can deliver material efficiency gains under realistic operating conditions, offering a practical route to greener, more reliable cloud services for Botswana’s public-sector and commercial workloads and a policy-relevant lever for demand-side efficiency in emerging data-centre deployments.