Timing issues in hard real-time / embedded systems

点击翻译按钮获取中文摘要

In many real-time applications, timing requirements are as important as any other requirements. Consider a medical device controlled by software that is designed to deliver a specific dosage of radiation. If the device delivers the correct amount of radiation, but for too long a time period, it is dangerous. The length of time over which the patient is irradiated is just as important as the level of radiation emitted by the device. These systems are known as hard real-time systems in that they have definite (hard) timing deadlines.     In a system such as mentioned above, there is an obvious safety concern. The software has to be developed so that it is provably safe. One step in the development of safety-critical software is the formulation of comprehensive requirements. Preferably, these requirements should be described with mathematical precision. A number of approaches have been developed to document timing requirements for hard real-time systems that are mathematically sound. However, these well-known approaches have significant deficiencies when they are used in real-world projects. For example, they do not cope adequately with tolerances.  The mathematical basis for some of these methods  can be shown to be invalid in the presence of tolerances.  Also, the intervals that result may be so large that the requirement becomes meaningless.     I was a key developer of the methodology used by Ontario Power Generation (OPG) to create safety-critical software for Ontario's nuclear plants. Together with colleagues and students, over the past four years we have formalized and extended the work I started at OPG, so that we can now describe typical timing requirements, and also predict the feasibility of their implementations - even when the requirements and implementations include tolerances. One particularly noteworthy result is that we have shown that longer sample intervals can achieve feasible implementations even when some shorter sample intervals cannot. My proposed research builds on this foundation. A primary goal is to produce software tools that help software professionals use these methods to build correct, precisely documented software applications that result in provably safe systems.