Hybrid Latency Minimization Approach using Model Checking and Dataflow Analysis

Abstract

Bounding the latency of real-time multiprocessor applications is crucial for safety-critical systems. Several approximative analysis approaches exist that can efficiently analyze the latency. However, these approaches produce pessimistic latency results and do not exploit buffer sizing nor exploit additional sequence constraints to reduce the latency. More accurate latency analysis results can be obtained using model checking of timed-automata, however, at the cost of a typically excessive run-time. This paper presents a latency analysis approach for cyclic task graphs using model checking of timed automata of which the run-time is reduced. The approach is applicable for systems in which tasks are executed on shared processors using a Fixed Priority Pre-emptive (FPP) scheduling policy. The reduction in run-time is achieved by pruning the search space of options that need to be analyzed using the model checker by making use of approximative dataflow analysis techniques. The approach exploits dimensioning of buffers to minimize interference and latency. Moreover, sequence constraints are introduced and automatically adapted in order to minimize the latency of the task graph. A WLAN 802.11p transceiver application is used in the case study to compare this hybrid analysis approach to a state-of-the-art approximation based approach that uses iterative buffer sizing. Using our approach, the analyzed latency decreased from 17 μs to 15 μs at the cost of a run-time of 23 minutes instead of a fraction of a second.