Deep Reinforcement and Meta-Heuristic Hybrid Models for Intelligent Cloud Load Distribution

bstract : Cloud computing has become the backbone of the
modern digital infrastructure, making the provision of on
demand, scalable and cost-efficient computing services possible. However, efficient load distribution is a major challenge because of unpredictable workloads and dynamic resource availability and service level agreement (SLA) limitations. Traditional static or heuristic-based schedulers cannot adapt well to a highly dynamic nature of cloud environments. This study proposes a Deep Reinforcement and Meta-Heuristic Hybrid Framework (DRMHF) aimed at getting intelligent, adaptive and energy efficient cloud load distribution. The framework combines adaptive decision making through Deep Reinforcement Learning (DRL) and global search refinement through meta-heuristic optimization technique, Ant Lion Optimization (ALO). Experiments performed on CloudSim Plus based on Google Cluster workload traces show significant performance improvements compared to the existing schedulers. The proposed DRMHF achieves a makepan of 1250 s, throughput of 820 task/s, energy consumption of 52.4 kWh and only 1.8% SLA violation,
which outperforms state-of-the-art approaches. These results
validate the hypothesis that hybrid intelligence can play an
important role in improving adaptability, scalability, and
sustainability in cloud load management, and determine DRMHF
as a promising strategy for autonomous resource allocation for large-scale cloud infrastructures.