Luca Abeni, University of Trento
Bernardo Villalba Frias, University of Trento
Luigi Palopoli, University of Trento
Daniele Fontanelli, University of Trento

An interesting trend in the design and development of modern embedded control systems is the increase of the CPU (and sensors) power, which leads to an increased resource sharing, and encourages to run multiple control applications on the same embedded board. However, running multiple applications on the same CPU/core reduces the determinism of the applications' response times (due to the interference between different applications), and the advanced CPU architectures tend to reduce the determinism of their execution times. These are critical issues given that real-time control applications must be highly predictable in terms of performance.

Under these conditions, the classical hard real-time design approach which allocate computational resources based on the worst-case demand of the application, results in overly conservative choices which could lead to a massive waste of resources and to a drastic reduction in resource sharing. These issues can be addressed by combining a proper CPU scheduler (based on resource reservations) with an task appropriate model (named "continuous stream task model") which allows to use a suitable stochastic description of the delay evolution in the control computation.

Since a reservation-based CPU scheduler (SCHED_DEADLINE) that can be used for this purpose has just been merged in the Linux kernel, Linux can now be used to implement the continuous stream task model described above. This paper presents an implementation (based on Linux and SCHED_DEADLINE) and experimental evaluation of these solutions in the context of a Linux-based robotic application, consisting in a mobile robot with the goal of following a line on the ground.

The analysis of the data collected from the real experiments confirms the consistency between theory and simulations and the experimental results, verifying the validity and effectiveness of the continuous stream approach in a real scenario. Furthermore, the presented experiments show that SCHED_DEADLINE can offer important advantages such as minimizing the resource utilization and allowing to provide guaranteed control performance.