Performance Analysis of Age of Information in a Foundation Model Native Infrastructure

The current processing capabilities and machine learning algorithms available at the network edge have ushered in a new era of intelligent services and applications. In machine learning tasks, the status information embedded in packets remains concealed until it undergoes complex data processing, which is both computationally demanding and time-consuming. In this context, the paper explores multiple sources sending individual process update data to a single network node, utilizing foundation models for efficient general-purpose time-series prediction. This approach eliminates the overhead and costs associated with training models from scratch while addressing the challenge of data scarcity. In particular, we formulate an Age of Information (AoI) problem and derive results based on classical AoI metrics proposing a simplified analytical approach, whose accuracy is validated through comparisons with simulation results of the actual system. Finally, we cast the novel analysis provided to perform time series forecasting using foundation models in relation to independent process behaviors.