Queuing Models for Field Defect Resolution Process

Abstract

This paper explores a novel application of queuing theory to the corrective software maintenance problem to support quantitative balancing between resources and responsiveness. Initially, we provide a detailed description of the states a defect traverses from find to fix and a definition and justification of mean time to resolution as a useful process metric. We consider the effect of queuing system structures, priority levels and priority disciplines on the differential mean times to resolution of defects of different severities. We find that modeling the defect resolution capacity of a software engineering group as n identical M/M/1 servers provides a flexible and realistic approximation to the queuing behavior of four different organizations. We consider three queuing disciplines. Though purely preemptive and non-preemptive priority disciplines may be suited for other groups, our data was best fit by a mixed discipline, one in which only the most severe defects preempt ongoing service activities of lesser severities. We provide two examples of the utility of such a model: given the reasonable assumption that the most severe defects have the highest impact on reliability, we find that the reduction of the resolution time for these defects must come from changes reducing the service time. On the other hand the effect of additional engineering resources on the resolution time of less severe defects is easily computed and can be significant