Reliability measurement during software development

Measurement of software reliability was carried out during the development of data base software for a multi-sensor tracking system. Every run made during this project was scored as success or failure, and supporting data were collected on forms for further analysis. The failure ratio (number of failures per calendar interval divided by total number of runs) and failure rate (number of failures divided by CPU time for the interval) were found to be consistent measures, on a month-to-month basis as well as from module to module, and therefore considered valid indicators of reliability in this environment. Trend lines could be established from these measurements that provide good visualization of the progress on the job as a whole as well as on individual modules. Over one-half of the observed failures were due to factors associated with the specific run submission rather than with the code proper.