Abstract:
This paper classifies software researches as theoretical researches，experimental researches，and engineering researches，and is mainly concerned with the experimental researches with focus on software reliability experimentation and contro1．The state—of-the-art of experimental or empirical studies is reviewed．A new experimentation methodology is proposed，which is largely theory discovering oriented．Several unexpected results of experimental studies are presented to justify the importance of software reliability experimentation and contro1．Finally，a few topics that deserve future investigation are identified．
Keywords: software reliability，software experimentation，software reliability control，adaptive testing
Introduction:
Software researches can roughly be classified as theory inventing，theory discovering，theory testing，and theory customerizing (applicablizing)．This is shown in Fig．1．
The term “theory” should be widely interpreted．It refers to an argument．an assertion．a principle，a model，or a hypothesis of software techniques．Software processes or／and software systems／products． Theory inventing means that some
kind of theories is created by researchers with intuitive judgments to characterize，describe，or interpret imaginary scenarios or observed phenomena．This is mainly the focus of theoretical researches． A famous example is the Turing machine．Theory discovering means some kind of theories is discovered from observed data or evidence to model or interpret the underlying principles of the observed data or evidence．The controversial software science of Halstead is an example of discovered theory． Theory testing means that some kind of theories is tested against observed data or evidence under certain circumstances or scenarios．The claim that the number of defects remaining in a game-playing software system is less than 20 is a theory that should be validated before the software is released． Theory discovering and theory testing are mainly the focus of the experimental researches of software．Finally，theory customerizing means some kind of theories is customerized to the software processes or systems of concern and the applicability of the theories is examined．The observation that software fault tree analysis techniques can efficiently identify causes of critical software faults and effectively improve design of flight control software is a customerized theory with respect to flight control software． This is mainly the focus of engineering researches．
Software reliability researches are devoted to software reliability improvement and assurance and can be tracked back to the work of Hudson in 1967 which models the behavior of software defect introduction and removal as a birth and death Poisson process． Large amount of software reliability research has been carried out，which can be classified as theoretical researches experimental researches，and engineering researches accordingly． By software reliability experimentation we mean experimental researches of software reliability in which software experiments are conducted． According to Basilii．an experiment is a form of empirical study where the researcher has control over some of the conditions in which the study takes place and control over the independent variables being studied；an operation carried out under controlled conditions in order to test a hypothesis against observation．With the undoubted and growing importance of software reliability problems in various application areas，a few questions arise naturally：what is the state—of-the—art of software reliability researches in general，and software reliability experimentation in particular? How should software reliability experiments be conducted? How can software experimentation help to evaluate new software reliability techniques?
This paper is aimed to address the above questions．More specifically, the aim of this paper is four—fold．First，this paper will review the current status of software reliability researches with focus on software reliability experimentation．This will be done in Section 2．Second．this paper will propose a methodology for software reliability experimentation．which will be presented in Section 3 to guide how software reliability experiments should be conducted． Section 4 will present a few experimental results generated by the software reliability
experiments should be conducted． Section 4 will present a few experimental results generated by the software reliability experimentation methodology．The third aim of this paper is to review a new software reliability technique for soft—ware reliability assessment，which is obtained by treating software reliability assessment as an adaptive control problem． This will be accomplished in Section 5 with preliminary experimental results．Finally．this paper will identify a few topics of research potential for software reliability experimentation and contro1． This will be demonstrated in Section 6．Concluding remarks are contained in Section 7．
Model-Based Studies and Empirical Studies
The systematic researches on software reliability started with the recognition of the so-called software crisis in the late 1960s, of which software reliability problem was one of the major causes．In 1970s，the central themes of software reliability researches were software reliability modeling and software fault—tolerance．A large number of software reliability models were proposed. These models are aim ed to quantitatively evaluate software reliability and／or estimate the number of defects remaining in the software of concern． 0n the other hand, several approaches to software fault—tolerance were proposed，including the N—version programming and
the recovery block programming． In this period the notion of software reliability was mainly interpreted as a quantitative index that measures the probability of failure—free operations of software systems． In other words．software reliability was interpreted in a narrow sense．The field of software reliability largely resembles that of hardware reliability．
In 1980s researchers begun to examine which and what techniques can help to improve and assure software reliability in engineering practice． Consequently, several important consensuses were reached． First, although certain software reliability techniques are useful and can help to improve and assure software
上一页 [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] ... 下一页 >>