New scientific contributions to the prediction of the reliability of critical systems which based on imperfect debugging method and the increase of quality of service

This paper presents a new method by which it is possible to realistically predict the software reliability of critical systems. The main feature of this method is that it allows estimating the number of remaining critical faults in the software. The algorithm employs well-known methods such as Imperfect Debugging and it provides a more reliable prognosis than the methods conventionally used for this purpose. Furthermore, the new approach describes two processes of handling critical failures (one for detection and one for correction). The new algorithm also takes into account the socalled repair time, a measurement that is vitally important for a reliable prognosis. For use in the prediction model, it is mathematically described as a time function. As every programmer knows, it can be difficult to have even the simplest program run without faults. So-called software reliability models (SRM´s), based on stochastic and aiming to predict the reliability of both software and hardware, have been used since the 70´s. SRM´s rely on certain model assumptions some of which cannot be deemed realistic anymore. Hence, for today´s reliability engineering, these models are insufficient. At this point in time, though, there are hardly any methods that enable us to obtain predictions as to how the reliability of critical faults or the failure rate of critical systems behave over time. Currently, there is no mathematical model distinguishing between critical and non-critical faults, and only few models consider Imperfect Debugging (ID). The method presented here, however, is based on ID and it is able to distinguish between critical and non-critical software faults. Moreover, this new method employs a so-called Time-Delay and thus two new processes have to be designed. Mathematically, these processes describe the detection of faults and their correction, respectively. It is necessary to define appropriate distribution functions and to clearly state the requisite model assumption- .