Latency upper bound for data chains of real-time periodic tasks

Abstract

The inter-task communication in embedded real-time systems can be achieved using various patterns and be subject to different timing constraints. One of the most basic communication patterns encountered in today's automotive and aerospace software is the data chain. Each task of the chain reads data from the previous task and delivers the results of its computation to the next task. The data passing does not affect the execution of the tasks that are activated periodically at their own rates. As there is no task synchronization, a task does not wait for its predecessor data; it may execute with old data and get new data at its later release. From the design stage of embedded real-time systems, evaluating if data chains meet their timing requirements, such as the latency constraint, is of the highest importance. The trade-off between accuracy and complexity of the timing analysis is a critical element in the optimization process. In this paper, we consider data chains of real-time periodic tasks executed by a fixed-priority preemptive scheduler upon a single processor. We present a method for the worst-case latency calculation of periodic tasks' data chains. As the method has an exponential time complexity, we derive a polynomial-time upper bound. Evaluations carried out on an automotive benchmark demonstrate that the average bound overestimation is less than 10 percent of the actual value.

[Solution] End-to-end Scheduling of Real-time Task Pipelines on Multiprocessors

“…There are other research studies that explore the end-to-end timing analysis of task chains [6,32,40,41] in practical scenarios such as in drones [16], and in ROS [13,65]. Proposed scheduling algorithms based on these approaches rely on job release times [7,17].…”

“…Task pipelines, or cause-effect chains, have received increased attention in recent research work [17,20,32,40,41,60], partly because of their usage in well-known software packages such as ROS [13,65]. Although real-time task pipelines have long been studied [23,49,50], the application of constraints on a pipeline has received little attention [19].…”

MIMOS: A Deterministic Model for the Design and Update of Real-Time Systems

“…Computing end-to-end latency can be studied in terms of the release patterns of the nodes in a MIMOS model. For instance, for a set of periodic tasks communicating through a set of registers, which can be viewed as a special case of Extended TKPN, the problem has been explored in [17] which presents a worst-case analysis method with exponential time complexity. Also, a polynomial-time approach is provided for computing an upper bound.…”