SCHEDULING METHOD FOR REAL-TIME PROCESSES IN CYBERPHYSICAL SYSTEMS BASED ON ZERO COPY

Abstract

The article proposes a model and architecture of cyber-physical systems (CFS) that reflect their key properties, including hierarchical organization, the presence of closed control loops, the distribution of functions between high-level and low-level algorithms, as well as close interaction between computational and physical components. The developed typical KFS architecture integrates sensors, actuators, computing resources and software of different levels through well-defined information and control links and forms a basic class of systems for researching real-time problems.

The main attention is paid to the architecture of the implementation of the method based on ROS2, which includes the KFS control unit, the real-time scheduler, the ROS 2 task executor and the memory management subsystem with support for zero-copy access. The proposed scheduling algorithms take into account the DAG structure of applications and provide deterministic access to shared memory, which is critical for systems with hardware accelerators. The combination of priority processing of root tasks with the use of a LIFO queue for child tasks allows to reduce execution delays and correctly implement priority inheritance mechanisms in complex graphs of data flows. The use of the zero-copy allocator eliminates the repeated copying of data between the central processor and hardware accelerators, stabilizes time characteristics and reduces the load on the memory subsystem.

Experimental evaluation confirmed the effectiveness of the proposed approach, demonstrating a significant reduction of average and maximum delays, jitter, and the number of deadline violations compared to classical methods. In addition, an improvement in the quality of control was recorded, in particular, a reduction in over-regulation and an increase in resistance to external disturbances. The obtained results indicate the expediency of the integrated design of planning, control and memory management and confirm the practical suitability of the proposed approach for a wide class of KFS with hardware accelerators.