3.5 SOME PORTABLE MODELS

< Day Day Up >

I now intend to provide a small number of models that I believe can be used as the basis of a set of management information measures within an organization. These are my own personal choice and have resulted from applying the metrics derivation procedures described later in section 2. I have used them successfully in a number of organizations but let me stress, they are models, not measures, and they will need additional work before you can use them as metrics for your organization. Also, I do not guarantee universal applicability—all I can say is that they have worked for me, so far, in a number of environments.

For measuring reliability, or at a higher level the quality of a software product, I suggest a simple defect density measure, i.e., defects reported from the field or customer divided by the size of the product. If you need to compare applications that are used at different rates, say in terms of the number of site installations, then I suggest you build in a usage metric to the denominator, multiplying not adding. I prefer defect density to the common Mean Time Between Failure (MTBF) measure for two reasons. First, I am always wary of using means within software data sets which tend to be skewed so that the mean is not a suitable level indicator and, second, I have never found it easy to collect failure data for the environments I have worked in. One other minor point is that failures are the really visible defects, the showstoppers. Many users get very fed up with the less critical defects that do not cause the system to fail but which are, nonetheless, defects or faults.

It is also possible to classify defects according to their severity but one should beware of introducing sophistication before the basics are in place. Incidentally, defect, error or fault reports used in this measure should be reduced to include only validated and non-duplicated reports. That means that only defects that are found to be the result of genuine faults should be included in the reliability measure and that if a defect is reported by more than one source, i.e., user, it should be treated as a single defect from the point of view of this measure.

< Day Day Up >