An Analytical Investigation of Software Sensitivity to Operational Proole Changes

Many software quality measures depend on an operational proole | a description of the environment where the software will be used. Usually the operational proole takes the form of a probability density governing the random selection of program inputs. When testing is used to estimate software quality, the ideal case is one where tests are selected according to the same operational proole that will be seen after the program is deployed. But this is not always possible, so there has recently been an interest in quantifying the eeect of operational proole errors on software quality estimates. In this paper we present an analytical approach to this problem. Treating the case where there are nitely many possible program inputs, we investigate how the performance of software varies over diierent operational prooles. We devise a measure on the set of operational prooles that allows us ask what proportion of prooles satisses a given criterion (such as the criterion causing a program to exhibit a certain probability of failure). We use this measure to look at the proportion of operational prooles that cause the software to perform poorly.