Microservice application for forecasting and optimizing network operations for temporary high-density IT infrastructure.

Abstract

This work introduces a much-needed improvement to a custom enterprise application named SUMMIT by adding forecasting capabilities to help with network operations in temporary high-density IT (THDI) environments, such as large technology trade shows. These settings experience rapidly changing network demand over short periods, which makes reactive IT management insufficient.<br /> <br /> This study evaluates short-term time-series forecasting methods with ARIMA, recurrent neural networks with Long Short-Term Memory (LSTM), and a hybrid ARIMA-LSTM model, using telemetry metrics such as active Internet connections and Network Address Translation (NAT) translations.<br /> <br /> An experimental design compared forecasting performance using realistic operational data. Exploratory cross-event validation was conducted to examine generalization across trade show environments of different scale and geography. The most accurate model has been integrated into the SUMMIT architecture as a dedicated forecasting microservice to predict short-term network load.<br /> <br /> This work contributes by applying time-series forecasting to operational network management through a deployable system. By integrating predictive analytics into a monitoring platform, the project provides a practical framework for proactively managing temporary high-density trade show networks.<br /> <br /> By improving small projects, the cumulative effect can be very significant on an enterprise, particularly by simplifying its architecture and process. That is one of the reasons this project used the microservice approach: to reduce complexity by creating dedicated services for specific tasks. Keywords: time-series forecasting, LSTM, ARIMA, network telemetry, temporary IT infrastructure, microservices, web visualization, twelve-factor, enterprise application patterns