A local approach to the testing of real-time systems

Real-time systems are computer systems whose behaviour relies on physical time requirements. The violation of these requirements by some implementation may be uncomfortable (e.g. videoconferencing), or even catastrophic (e.g. process control), and thus damage the overall operation of the concerned systems. As more and more time-dependent products are becoming available on the market, the interest in testing them is growing, and standardization institutions (such as ISO/IEC) are working on the definition of formal description techniques for specifying real-time communication systems. But behaviour computation for real-time specifications generally involves exponential complexity. The work presented in this paper aims at avoiding this limitation in the test-development process. We use timed input-output automata to model systems behaviour. As implementations under test are 'black-boxes', the test verdict relies on the observation of behaviour that is inherent in some time intervals. These time intervals that characterize the success or the failure of a test experiment are computed in a local manner, from the synchronous product of a given specification and a given test requirement. An example with the ATM ABR protocol is provided to illustrate the work.