Reliability Analysis of Real-Time Multitasking Software Based on Neural Network - Advanced Research on Computer Education, Simulation and Modeling - page 349

t > t

Time t obeys the distribution of t exe . If , (5) can be re-wrote as

rel

n

t + p ( t − t )

∑

∫

i i rel

R ( t ) = exp( ( − Φ ( t ) dt ))

. (6)

i

rel

i = 1

3 Reliability Model of Real-Time Multitasking Software

3.1 Neural Network Model of General Software Reliability

With method combined Halstead theory [2] and G-O model [3], we can know that
reliability index currently can be arrived according to software code line number B s ,
software test time T s , failure number in test N s and running time T . Good approxima-
tion capability of RBFNN can be used to establish its reliability analyzing model. The
parameters B s , T s . N s and T are input of RBFNN and reliability index is its output.
Here failure rate Φ is selected. Structure of RBFNN is shown in Fig. 1. Failure rate
can be used to describe software reliability. RBFNN was trained with actual measured
data. After the success of network training, the trained network can be used to analyze
its reliability index with some new specific inputs.

B s

Output failure rate Φ

T s

i

∑

N s

…

T

Fig. 1. Neural network model of general software reliability

3.2 Neural Network Model of Real-Time Multitasking Software Reliability

Each task module of real-time multitasking software has relatively independent func-
tions. Real-time was scheduled by operation system with some scheduling strategy to
occupy CPU resource. Each task module has its own independent reliability. Based on
structural and modular idea of Littlewood, we can independently analyze on reliabil-
ity of each task module, and then to draw reliability of whole software. For difference
of running period and execution time, the proportion P i of each module i ( i =1,2,…, n )
account for overall running time is different. Task module that occupies larger propor-
tion of overall running time may have greater impact on system reliability. So P i is an
important factor to determine software reliability [4].

P = C / t

, where C i is task

i i i