Optimizing Preventive Service of Software Products

Abstract

The implementer of a large, complex software system cannot make it completely defect free, so he must normally provide fixes for defects found after the code is put into service. A system user may do preventive service by installing these fixes before the defects cause him problems. Preventive service can benefit both the software developer and the software user to the extent that it reduces the number of operational problems caused by software errors, but it requires the expenditure of the resources required to prepare, disseminate, and install fixes; and it can be the cause of additional software problems caused by design errors introduced into the code by fixes. The benefit from removing a given defect depends on how many problems it would otherwise cause. Benefits may be estimated by modeling problem occurrence as a random process in execution time governed by a distribution of characteristic rates. It is found that most of the benefit to be realized by preventive service comes from removing a relatively small number of high-rate defects that are found early in the service life of the code. For the typical user corrective service would seem preferable to preventive service as a way of dealing with most defects found after code has had some hundreds of months of usage.