Vous êtes sur la page 1sur 1

1188 World Abstracts on Microelectronics and Reliability

IEEE Trans. Reliab. R-30 (4), 325 (1981). In a probabilistic by analyzing an initial segment of the written code. Two
network, source-to-multiple-terminal reliability (SMT reli- parameters are estimated from the data. The data are: (a)
ability) is the probability that a specified vertex can reach time separations between error detections, (b) the number of
every other vertex. This paper derives a new topological errors per written instruction, (c) the failure rate (or finding
formula for the S M T reliability of probabilistic networks. rate) of a single error, and (d) a time record of the number of
The formula generates only non-cancelling terms. The non- instructions under test. This model permits predictions of
cancelling terms in the reliability expression correspond one- M T T F and error content of any software package which is
to-one with the acyclic t-subgraphs of the network. An homogenous with respect to its complexity (error making/
acyclic t-subgraph is an acyclic graph in which every link is in finding). It assists in determining the quality, as measured by
at least one spanning rooted tree of the graph. The sign to be error contents, early on, and could eliminate the present
associated with each term is easily computed by counting the practice of applying models to the wrong regimes (decreasing
vertices and links in the corresponding subgraph. failure rate models applied to growing-in-size software pack-
Overall reliability is the probability that every vertex can ages). The growth model is very tractable analytically. The
reach every other vertex in the network. For an undirected important requirement for applications is that the error-
network, it is shown the S M T reliability is equal to the making rate must be constant across the entire software
overall reliability. The formula is general and applies to net- program.
works containing directed or undirected links. Furthermore
link failures in the network can be s-dependent. A heuristic algorithm for system failure frequency. J. P.
An algorithm is presented for generating all acyclic t-sub- GADANI and K. B. MISRA. IEEE Trans. Reliab. R-30 (4), 357
graphs and computing the reliability of the network. The (1981). For reliability analysis, the choice between tie-sets
reliability expression is obtained in symbolic factored form. and cut-sets is not a matter of convenience but depends on
assessment of computational effort. The system configura-
Rules for calculating the time-specific frequency of system tion alone determines the preference of the approach. The
failure. C. SINGH. IEEE Trans. Reliab. R-30 (4), 364 (1981). present paper offers a faster algorithm for evaluating system
This paper develops formulas for converting expressions for failure frequency using both approaches. An expression for
time-specific probabilities of system success and system system failure frequency between two specified nodes is de-
failure into time-specific fi'equencies of system success and rived using either tie-sets or cut-sets. The expression is in
system failure. terms of failure frequency, availability, and unavailability of
the system elements. A heuristic algorithm to obtain the
Stochastic reliability-growth: a model for fault-removal in expression can be programmed for a computer. The algo-
computer-programs and hardware-designs. BEV LITTLEWOOD. rithm is illustrated through an example. The manner in
IEEE Trans. Reliab. R-30 (4), 313 (1981). An assumption which Abraham's algorithm can be used to determine system
commonly made in early models ofsoftware reliability is that probability of failure using cut-sets is suggested. The algo-
the failure rate of a program is a constant multiple of the rithms are also useful for evaluating sensitivity coefficients.
(unknown) number of faults remaining. This implies that all
faults contribute the same a m o u n t to the failure rate of the Availability, MTBF and MTTR for repairable m out of n
program. The assumption is challenged and an alternative system. PER A. KULLSTAM.IEEE Trans. Reliab. R-30 14), 393
proposed. The suggested model results in earlier fault-fixes (1981). The availability of a redundant system requiring m
having a greater effect than later ones (the faults which make out of n units to be operational represents a basic and ele-
the greatest contribution to the overall failure rate tend to mentary result within reliability engineering. The corres-
show themselves earlier, and so are fixed earlier), and the ponding mean time between system failures (MTBF) and
D F R property between fault fixes (assurance about pro- mean time of system restoral (MTTR) are less commonly
grams increases during periods of failure-free operation, as known since these results have either required the solution of
well as at fault fixes). The model is tractable and allows a time-dependent system of equations or matrix algebra. This
variety of reliability measures to be calculated. Predictions of short paper shows that explicitly results for the M T B F and
total execution time to achieve a target reliability, and total M T T R performance are as easy to obtain as the availability
number of fault fixes to target reliability, are obtained. The for arbitrary m, n-values.
model might also apply to hardware reliability growth result-
ing from the elimination of design errors. Limitations of minimal cut-set approach in evaluating reli-
ability of systems with repairable components. C.A.
A useful property of "Graphical" estimators of location and CLAROTTI. IEEE Trans. Reliab. R-30 (4), 335 (1981). This
scale parameters. PASQUALEERTO. IEEE Trans. Reliab. R-30 work shows that, when system components are repairable,
(4), 381 (1981). This paper presents a method for construc- system reliability cannot be evaluated exactly by means of
ting exact s-confidence limits for reliability and/or quantiles the fault tree approach; only a system reliability lower bound
based on graphical estimators of location and scale para- can be obtained.
meters. The method arises from the ancillary property of the
reliability function estimated by using plotting positions and Use of Q-GERT network simulation in reliability analysis.
the least squares method. From the distribution of this esti- EDWARD R. CLAYTON and JOHN W. COOLEY. IEEE Trans.
mator alone, one can obtain graphs or tables to calculate Reliab. R-30 (4), 321 (1981). Reliability analysis for all but the
directly the s-confidence limits for both reliability and quan- simplest systems often is unmanageable. This paper proposes
tiles. These limits are exact even for very small sample-sizes Q - G E R T simulation for such analysis. Q - G E R T as opposed
in that they are calculated without using asymptotic approxi- to the original G E R T analysis, is a relatively easier network-
mations. An example of a graph which enables one to cal- ing model. Thus, determining the reliability of both simple
culate these limits for a 2-parameter Weibull distribution, is and complex systems from individual component test data
given. This example uses plotting positions recently pro- lends itself quite readily to Q - G E R T analysis. Although the
posed by the author. examples are rather simple, mathematically tractable
systems, it is possible to extend Q - G E R T procedures to more
An error detection model for application during software de- complex systems. The probability distributions for these
velopment. PAUL B. MORANDA.IEEE Trans. Reliab. R-30 (4), more complex systems may be either user defined or may be
309 (1981). A variation of the Jelinski/Moranda model is chosen from any one of l l different predefined functions. The
described. The main feature of this new model is that the only requirement for the use of Q - G E R T is that input data
variable (growing) size of a developing program is accom- describing the model be provided by the analyst. The basic
modated, so that the quality of a program can be estimated format of the Q - G E R T model facilitates this description of

Vous aimerez peut-être aussi