Partially Stateful Server Selection for Minimal Age of Information Scheduling Over a Finite Horizon

We consider a source reporting real-time information content over a finite horizon so as to obtain minimal age of information (AoI). We assume that the information content requires a computationally heavy handling, as typical of tasks involving AI-augmented interpretation. As such, scheduling an information update requires a processing can be either performed locally, or offloaded to a remote mobile edge computing (MEC) server shared by other sources. The former option is subject to a certain failure rate, whereas the latter is always successful, but taking a longer time subject to how many other similar sources use the remote server. Inspired by the literature for MEC offloading, we consider a partially stateful approach, where the scheduling decision is made according to the system state (comprising the current AoI, the number of updates available, and the current congestion at the remote server), whereas the server selection follows a randomized-alpha policy. Through a dynamic programming approach, we find that the optimal choice of the local processing rate, although dependent on the characteristics of the remote server, is relatively robust to its variations. Not only does this justify our approach, but it also highlight a practical low-complexity approach to draw meaningful considerations on server sharing in multi-source updating systems.