What is software reliability?

Abstract

Reliability refers to statistical measures an engineer uses to quantify imperfection in practice. Often we speak imprecisely of an object having "high reliability", but technically, unless the object cannot fail at all, its reliability is arbitrarily close to zero for a long enough period of operation. This is merely an expression of the truism that an imperfect object must eventually fail. At first sight, it seems that software should have a sensible reliability, as other engineered objects do. But the application of the usual mathematics is not justified. Reliability theory applies to random (as opposed to systematic) variations in a population of similar objects, whereas software defects are all design flaws, not at all random, in a unique object. The traditional cause of failure is a random process of wear and tear, while software is forever as good (or as bad!) as new. However, software defects can be thought of as lurking in wait for the user requests that excite them, like a minefield through which the user must walk.<>