Theoretical Feasibility of Statistical Assurance of Programmable Systems Based on Simulation Tests

This presents a new model to support empirical failure probability estimation for a software-intensive system. The new element of the approach is that it combines the results of testing using a simulated hardware platform with results from testing on the real platform. This approach addresses a serious practical limitation of a technique known as statistical testing. This limitation will be called the test time expansion problem (or simply the 'time problem'), which is that the amount of testing required to demonstrate useful levels of reliability over a time period T is many orders of magnitude greater than T. The time problem arises whether the aim is to demonstrate ultra-high reliability levels for protection system, or to demonstrate any (desirable) reliability levels for continuous operation ('high demand') systems. Specifically, the theoretical feasibility of a platform simulation approach is considered since, if this is not proven, questions of practical implementation are moot. Subject to the assumptions made in the paper, theoretical feasibility is demonstrated.