Computer Software Reliability: Many-State Markov Modeling Techniques.

Many-state Markov models have been developed for the purpose of providing quantitative reliability criteria for computer software. The software system under consideration is assumed to be large, so that statistical deductions become meaningful, and is assumed to initially contain an unknown number of bugs. The basic models provide estimates and predictions for a quantifier that represents the state of debugging of the system and which is generally the most probable number of software errors that will have been corrected at a given time in the operation of this software system based upon preliminary modeling of the error occurrence rate and the error correction rate. The models also provide predictions for the availability and for the reliability of the system. The differential equations corresponding to the basic many-state Markov models are solved for verification and demonstrative purposes. Manufacturers data have been obtained on this performance of system software for a medium-sized software operating system. These data have been analyzed to obtain frequency distributions of the random variables representating the time to close software error reports. The data are then used for application of the basic many-state Markov model. A general discussion of error data collection is undertaken in some detail, and suggestions are made for possible improvements in software error data documentation practices.