Model Agnostic Predictive Smart Load Balancer in Microservices Environment

This paper presents a model-agnostic predictive smart load balancer for microservices environments, utilizing machine learning to optimize load distribution and resource utilization. The system leverages access logs from NASA, Calgary, and Clarknet datasets to categorize requests and estimate service times for various request types. Advanced ML models, including LSTM, GRU, RNN, ARIMA, and SARIMA, are employed to predict future loads on microservice instances. The load balancer uses these predictions to make proactive decisions, redirecting requests using a weighted round-robin algorithm and preemptively creating new instances when necessary. This approach ensures efficient handling of incoming requests, reduces wait times, and prevents resource over-provisioning. The system’s model-agnostic design allows for seamless integration of new ML models, enhancing its flexibility and robustness. Experimental results demonstrate significant improvements in load management and cost optimization compared to traditional load balancers, highlighting the effectiveness of incorporating ML for intelligent load balancing.