Embedded Reconfiguration of TSN: Dual Reconfiguration with Dropping and Reclaiming

Abstract

This paper presents our solution to the Industrial Challenge on embedded reconfiguration of Time-Sensitive Networking (TSN), held at the 37-th ECRTS. The challenge requires restoring, at runtime and without precomputed solutions, the schedules of streams affected by a link failure in a realistic avionics network. Our solution applies a DROP policy at all bridges to purge the affected streams, updates Gate Control Lists (GCLs) accordingly, reclaims the transmission windows held by the dropped streams on operational links, and applies incremental scheduling to restore as many streams as possible in priority order. Transmission windows of the rescheduled streams are fit into the available gaps without modifying the schedule of unaffected streams. After the incremental scheduling stage, a last-resort mechanism purges all lower-priority streams than any yet-unscheduled critical stream, reclaims their resources, and retries; sacrificed streams are subsequently tested for rescheduling. Evaluated on the benchmark of the challenge - 5 bridges, 15 end-stations, and 241 streams across eight criticality classes - the system achieves perfect recovery in all eight single-link failure scenarios, with total reconfiguration times below 4.4 s. In a harder scenario, reducing the topology to a linear chain through four simultaneous link failures, all safety-critical streams are still recovered. Activating the last-resort mechanism recovers six additional lower-priority streams, with 96.9% of sacrificed streams subsequently re-rescheduled. This paper extends our original challenge submission with a formal algorithm description, a comprehensive experimental evaluation using Gurobi, and the priority-driven last-resort mechanism.