Modeling time-triggered service intermittence in network calculus

Abstract

Network elements (e.g., switches), which intermit service, i.e., stop forwarding and transmission of data according to a repeating schedule, can be found in many real-time capable communication networks, e.g., communication networks with TDMA, Ethernet with Time-aware Shapers or low-power wireless networks. The behavior of those network elements depends on the (stationary) properties of the network elements, their schedule, and the current time, as well as the offered traffic load. If a networked real-time system generates traffic flows which are not synchronized to the schedules of the network elements, formal frameworks such as Network Calculus (NC) are highly valuable to derive deterministic guarantees for the communication. In this paper, we show the fundamental implications of modeling time-triggered network elements with service intermittence in NC. We identify two archetypes of network elements with intermittent service, and propose time-variant and time-invariant approaches to derive service curve formulations to model them. We evaluate the differences between time-variant and time-invariant service curves with respect to the overestimation of worst-case backlog and worst-case delay, and we identify schedule properties which influence the tightness of the derived bounds.