Abstract
The rapid development of time-sensitive applications, such as industrial automation, healthcare, and the Internet of Things (IoT), has amplified the demand for low-latency networks, particularly in the context of fifth-generation (5G) mobile networks. With stringent latency requirements es-sential for the smooth operation of these applications, the ability to predict and manage latency becomes critical. This research investigates and compares two forecasting models, seasonal auto-regressive integrated moving average (SARIMA) and error, trend, and seasonal (ETS), for pre-dicting network latency in internet protocol/multi-protocol label switching (IP/MPLS) networks, particularly in 5G environments where latency management is crucial for maintaining high-quality services.
The study evaluates both models by applying them to latency data from a 10-gigabit per second (Gbps) IP/MPLS fiber ring network. The data, collected over a period of 15 days, includes six different network links, with the study comparing latency performance across two aggregated routes, Route 1 and Route 2. Forecasting accuracy is assessed using several performance metrics. The results indicate that while SARIMA performs well in capturing the seasonal patterns of latency in more stable routes like Route 1, it struggles with the volatility observed in Route 2. ETS, while computationally more efficient, demonstrated a robust performance for both routes but showed limitations in handling the erratic behavior of Route 2.
This research underscores the importance of selecting an appropriate forecasting model based on the data's complexity and volatility. The findings suggest that SARIMA is more suited for net-works with stable latency patterns, while ETS provides a faster and more resource-efficient solu-tion for environments with less variability. The results contribute to the growing body of knowledge on 5G network management, offering insights into effective latency forecasting tech-niques that can enhance the efficiency and reliability of modern network infrastructures.
Future work should explore hybrid models that combine the strengths of SARIMA and ETS, po-tentially improving both forecasting accuracy and computational efficiency. Moreover, integrating machine learning (ML) techniques could further enhance the adaptability of forecasting models for managing latency in dynamic 5G networks.