Research On Software Reliability Growth Models Based On Queuing Theory

Software reliability is recognized as one of the most characters of software quality,and it relate to whether computer program or larger system operate in a specified environ-ment for a specified time. To estimate and predict software reliability, an important ap-proach is to apply the stochastic model called software reliability growth model (SRGM).Recently, researchers have tried to use server queuing models to describe software testingbehavior. However, fault detection efort (FDE) and fault correction efort (FCE), removalwaiting delay (RWD) and fault correction delay, and the influence of software reliabilityare not still addressed. It reduces the practical significance and the credibility of models Inorder to meet the practical requirement to estimate and predict reliability during softwaretesting, a series of SRGMs based on server queuing are studied in this dissertation.During the software testing phase, software reliability is highly related to the testefort. Compared with calendar time prediction, the time distribution of the testing eforthas the greatest efect on the shape of the observed software reliability growth curve. Faultdetection and fault correction are usually performed by diferent people during the soft-ware system test phase. In this case, resources spent on fault detection and fault correctionprocesses are considered as separate. This dissertation defines fault detection efort andfault correction efort as the testing efort consumption spent on the fault detection andcorrection processes, respectively. Namely, testing efort is composed of detection efortand correction efort.One major assumption commonly concerned with the SRGMs is that each time afault occurs, the fault that caused the failure is immediately corrected. This assump-tion may not be reasonable. In reality, there is often a considerable time delay betweendetecting a failure and correcting it, and the delay depends on the FCE, the number offault detected, the complexity of fault and program size and so on. This dissertationwill be concerned the issue of the delay and studied how to use Infinite Server Queuing(ISQ) model to describe software fault detection and correction processes behavior. In thequeuing-based approaches of modeling software reliability, detected faults and debugger-s are usually regarded as customers and servers, respectively. Based on queuing-basedmodels, The influences of FDE and FCE on the delay are considered, and ISQ model-s with FDE and FCE are proposed in this dissertation, i.e. ISQ-FDEFCE models. In addition, ISQ-FDEFCE models which correcting faults create another are proposed.Most SRGMs assume that the rate of fault correction follows the same distributionduring the fault correction process. However, in more realistic situations, the rate of faultcorrection strongly depends on FCE, the difculty of the problems debugging environ-ment and tools,etc. Once these factors are changed during the fault correction process,the rate of fault correction may be changed. Thus, the rate of fault correction may be nei-ther constant nor smooth. Instead, it may be changed at certain moments in time, whichis called as a change-point (CP) problem. Based on the foregoing discussion, in this dis-sertation, the CP problems will be studied in fault correction process, and ISQ-FDEFCEmodels based on the technique of CP are proposed, i.e. ISQ-FDEFCE-CP models. Inaddition, ISQ-FDEFCE-CP models whic correcting faults create another are proposed.ISQ models possess infinite resources to detect and correct every software fault. Thenew detected faults are corrected immediately and need not be permitted a period of wait-ing prior to correction. The time delay between the fault detection and fault correctionprocesses is fault correction delay, i.e. the waiting time to remove a fault is negligible.This assumption may not be reasonable and may not be always occur in real world. Thus,the assumption of ISQ models should be modified. In this dissertation, finite server queu-ing (FSQ) models are used to estimate and predict software reliability, considering finiteresources. The FSQ models based on removal waiting delay called FSQ-RWD are built.In addition, FSQ-RWD models which correcting faults create another are proposed.With the proceeding of the test, the software faults are corrected and the rate offailures will be gradually decreased. The speed of fault to enter the state waiting to becorrected will be reduced and the servers may be idle. The influences of adjustment ofthe number of server on removal waiting delay and fault correcting delay are studied insoftware fault correction process. For the adjustment of the number of server, FSQ-RWDmodels based on the technique of CP are built, i.e. FSQ-RWD-CP models. In additionFSQ-RWD-CP models which correcting faults create another are proposed. Experimentresults based on real failure data show the proposed models in this dissertation can givemore accuracy result in the data set. These models can help software companies to makebetter use of the SRGMs for supervision and management of software quality.