An essential part of the development of real-time software is the process of verifying that the software will meet its timing constraints. This paper presents a new set of techniques for a systematic worst-case schedulability analysis of real-time programs. These techniques utilize information about the program’s behavior, collected from the source code, and about the hardware. As the worst-case behavior must be predictable, the software and hardware must meet some restrictions, including having no constructs with arbitrarily long execution times (like while loops), having only some restricted forms of paging, and having “no” cycle stealing. On the other hand, real-time systems satisfying the restrictions can be analyzed automatically and with greater accuracy than older methods seem to be capable of.

The paper describes the techniques, a prototype implementation, and evaluation tests with two distributed real-time programs (of “modest size,” that is, less than 5,000 lines of code). The results in these cases show a clear improvement in the prediction accuracy as compared with older methods.

A couple of problems probably restrict the usability of these techniques. One problem is their computational complexity. Another is associated with the restrictions on the system: in realistic applications, these restrictions may turn out to be unacceptable. On the other hand, these problems indicate characteristics that may prevent any systematic schedulability analysis.

The paper is well written, and the techniques present an interesting approach to the schedulability analysis. The paper also gives the interested reader some insight into the performance structure of real-time programs.