Dynamic Load Management in High Traffic Web Applications Using Machine Learning

The paper proposes a framework to handle the web traffic for efficient resource utilization by employing token generation and waiting room strategies for managing user load during peak hours. The framework monitors active user sessions and active traffic queues using predictive modeling techniques based on Logistic Regression resulting in faster server response times and higher customer satisfaction. The main aim of this approach is managing zenith states in server utilization by restricting users from accessing the server simultaneously. A customer is given a token on entering the system, which allows monitoring of the number of active sessions. This strategy makes it impossible to overload the server and enhances the user's experience as time spent on the server is reduced. Predictive modeling through logistic regression using historical logs in the framework is a key aspect of predicting load. The model makes predictions incremental by borrow influences from zeitgeber and user systems to expect traffic increments and tasks resources appropriately. Dynamic plots in Matlab depict the systems' targeting concurrent users transformations as well as the system's response times.