Dynamic Deterministic Quality of Service Model with Behavior-Adaptive Latency Bounds
Robin Laidig, Frank Dürr, Kurt Rothermel, Stefan Wildhagen, Frank Allgöwer · 2023
Many time-sensitive networked systems, such as networked control systems or other cyber/physical systems, require well-defined Quality of Service (QoS) with guaranteed deterministic bounds on network delay. Existing QoS models typically provide no guarantees for excess traffic beyond the traffic specified during the initial admission process. This can lead to a waste of resources when applications overcompensate during resource reservation to avoid traffic violations. In this work, we propose the Dynamic Priority Token Bucket (DPTB), a fundamentally different new QoS model. DPTB permits short-time violations without immediately dropping to best-effort guarantees for excess traffic. Instead, the priority of the application gets degraded to weaker but still deterministic guarantees. At any point in time, the application can calculate the currently guaranteed delay bounds, which depend only on its own past sending behavior, to enable application-level adaptation of the sending rate and application-side prediction of the implications onto the application performance. We designed DPTB as a token bucket extension that can be used on top of several existing scheduling mechanisms, such as the Asynchronous Traffic Shaper of IEEE Time-Sensitive Networking (TSN). Our evaluations show that DPTB is more resilient to bursty cross-traffic, resulting in significantly lower average delays than regular reservations.