A Systematic Evaluation of MPSoC-enabled Real-Time Containers for Space Environment Applications

Abstract

In recent years, the growing number of satellites has motivated space agencies to open spacecraft software development to third parties, highlighting the need for ways to abstract away the complexity linked to the aerospace domain, in order to speed up the development of experiments that will run in space. For this reason, containerization is emerging as a promising solution to simplify deployment and provide isolation among applications, but its compatibility with real-time requirements - essential in space applications - remains largely unexplored, especially on embedded ARM platforms used in space agencies like DLR.

This thesis investigates the feasibility of enabling real-time containers on MPSoC-based boards representative of on-board computers used in space. A complete software stack was designed and implemented, combining a Linux kernel extended with Xenomai 4’s EVL real-time core and container-based isolation provided by Podman. The system was frst implemented on an emulated ARM Cortex-A53 Zynq UltraScale+ MPSoC platform using PetaLinux and QEMU, and then it was also deployed on the real (same) ZCU102 board.

An extensive experimental evaluation has been conducted to assess three main aspects: the ability to run real-time applications inside containers, their concurrent execution, and the scalability of the system on the target hardware. The results have shown that real-time execution within containers is feasible, although it introduces overhead and reduces determinism compared to bare-metal execution, and that multiple concurrent containers are also able to run while guaranteeing real-time capabilities for the applications inside them. An interesting aspect has been the feasibility of executing rootless containers with real-time capabilities, since they are more valued for safety reasons.

Overall, this work demonstrates that real-time containers on embedded platforms are a viable solution for space-oriented applications where isolation, performance and predictability are the most important guarantees to achieve.