Machine Learning-Optimized Load Balancing Algorithm for Indian Data Centers

This paper has tried to develop a machine learning-optimized load balancing algorithm that would apply specifically to an Indian data center with the expectation of improving performance, efficiency, and scalability. The method consists of Multi-Agent Deep Reinforcement Learning (MADRL) to address the decentralized decision-making problem, Federated LSTM Forecasting to produce quality and privacy-preserving workload forecasting and Digital Twin Simulation to simulate and test the system learning in a realistic virtual environment. The system is implemented based on the Ray RLlib framework that allows the intelligent distribution of tasks of geographically distributed data centers and works with local energy allocation and data privacy laws. The achieved result using the experimental assessment shows substantial benefits such as a decrease in response time, increase in CPU usage, low power consumption, and fewer SLA violations inclusive of the standard approaches Round Robin and Static Heuristics. The combination of federated learning will make sure that the data is localized, and the digital twin to facilitate low-risk, low-cost testing. The findings point towards the success of the proposed framework as an extendable and flexible approach towards new-age Indian data center systems.