Predictable Low-Latency Interrupt Response with General-Purpose Systems

Real-time applications require predictable and often low-latency response times when reacting to external events. Realtime operating systems allow applications to meet such timing requirements, but they offer less functionality and fewer APIs than a general-purpose operating system such as Linux. In this work, we present a virtualization layer that enables predictable, lowlatency interrupt handling for Linux-based real-time applications, even if Linux itself or unrelated applications cause heavy load. The benefit of our approach is that developers can draw from the rich feature set and powerful infrastructure provided by Linux, but their applications can meet timing requirements as if they ran directly on a real-time operating system. Our benchmarks show a reduction of worst-case interrupt response times by more than two orders of magnitude compared to standard Linux, and by a factor of 3 on average.