Approximate Edge-Server Placement Evaluation for Cyber-Physical Urban Rail Systems

Abstract

Trace-driven simulation captures mobility, communication, compute contention, and deadlines in edge-server placement studies, but repeated evaluation of candidate placements is costly. This paper compares static density-based proxies and sampled-window simulation with full-horizon simulation on 990 placements across three urban tram networks. Placement generation and static evaluation use tram residence-time density, while held-out workload realisations are reserved for simulation. The covered-density proxy achieved mean Kendall rank correlation 0.943 and retained the population-best placement within its top ten in all ten city–budget cells. Its median evaluation time was 13.5 ms per placement, compared with 72 s for full simulation. However, the proxy was not a calibrated quality-of-service (QoS) estimator and did not reliably preserve feasible placements under strict deadline-miss constraints. Sampled-window simulation reduced evaluation time and absolute QoS error, but its accuracy depended on the temporal position of the window under non-stationary demand. The results distinguish candidate preselection from QoS certification: static proxies can reduce the placement population, whereas simulation remains necessary to estimate QoS and evaluate constraints.