An Empirical Evaluation of a Method for Verifying Concurrent Java Components

1
An Empirical Evaluation of a Method for Verifying
Concurrent Java Components
Author Zahir Ally Supervisors Paul
Strooper Luke Wildman
AIM To empirically evaluate the ninth step of
the TestCon method and determine its necessity in
the overall method. EXPERIMENTAL DESIGN Object
of Study The TestCon method. A nine step method
using a combination of code inspection and static
and dynamic analysis tools to detect common
concurrency faults, such as interference,
starvation, deadlocks and livelocks. The final
step of the TestCon method requires the use of a
testing tool, named ConAn, which provides unit
testing in order to test the functional
requirements of the component. Purpose This
experiment is being performed to provide the
developers of the TestCon method with an
assurance that the final step of the method is a
useful addition to the overall method. Hypothesis
The numbers of unique errors detected by the
first eight steps of the TestCon method are less
than the number of errors detected by all nine
steps of the method. Population The experiment
will be applied to a thirty assignments developed
for the Concurrent and Real-Time Systems course
at the University of Queensland in 2002.
ANALYSIS AND INTERPRETATION Hypothesis Testing
The experiment is evaluated by the t-test
statistical test. We reject the null hypothesis
of Errors(Step1-8) Errors(Step1-9) and accept
the hypothesis of Errors(Step1-8) ?
Errors(Step1-9) if t0 gt t0.025, 60 ?
-2.566 gt 2.000 Summary As seen from the graph
below and the statistical analysis, the
experiment is able to show that there is a
significant improvement in the number of errors
detected when using all nine steps of the
TestCon method, as opposed to using the first
eight.
Figure 1. Comparative view of errors detected.