The ubiquity of the Linux kernel has made its application in real-time systems increasingly common. However, while recent advances in the kernel have reduced its interrupt latency dramatically, Linux's historic focus on throughput optimization with aggressive caching make it surprisingly difficult to guarantee latency bounds in "real-world" usage. This talk explains why and compares various approaches to software partitioning that facilitate the bounding of response latency, some of which can work without making any unusual demands on Linux itself.

This presentation will be much more about high-level architectural alternatives than coding details. The intended audience is systems engineers, technical management, and software developers seeking to use Linux or other general purpose OS for applications with hard real-time constraints.