3 Real-time operating systems

The difference between non-real-time operating systems like Windows and Linux and real-time operating systems like VxWorks is that real-time operating systems have short guaranteed thread preemption times and interrupt latencies. A standard Linux for example has interrupt latencies up to a few hundred milliseconds [11] while the latencies of current real-time operating systems are around tens of microseconds which are guaranteed by the system architecture. Worst-case times of a standard Linux can be determined by measurements but there is no guarantee that there are worse cases with even longer latencies.

Real-time operating systems are much more uncomfortable to program than desktop operating systems. They often do not provide a process concept (memory protection), which separates memory spaces of different applications and there is not much abstraction from the underlying hardware. Because applications running on embedded devices become more and more complex additional abstractions are needed. Mostly it is only a small part of a real-time application really having real-time requirements. For the rest of the software one would like to have the convenience of desktop operating systems. Therefore it makes sense to modify desktop operating systems to satisfy embedded and real time needs (or having a programming environment like Java, which we discuss later). To add real-time capabilities to a non-real-time operating system there are two approaches:

• Preemption Improvement: Modify the operating system to be preemptible in a defined, short time.
• Interrupt Abstraction: is a two kernel model running the unmodified operating system as the idle-task on top of a separate scheduler (also known as Interrupt Isolation or Interrupt Virtualization).

The fist approach is implemented for example by MontaVista [14], who improved the preemption times of Linux. Two kernel solutions for Linux are RTLinux [7] and RTAI [8]. VxWin [16] and VenturCom's RTX [17] are two-kernel solutions for Windows NT.

Due to the complexity of operating systems like Linux it is a very difficult task to improve preemption times. Because it is really hard to examine every code path in the kernel, worst case preemption times cannot be guaranteed. The interrupt abstraction model allows predicting worse case times in the small real-time layer exactly. A detailed comparison of the two approaches for Linux can be read in the articles [9] [10] [11].