Web Testing for Reliability Improvement - Advances in Computers - page 203

In our hierarchical strategy for web testing, information captured in our TAR and
CPR is similar to the stationary probability and transition probability above. Conse-
quently, we rely less on the last two thresholds above, and instead primarily use the
overall probability threshold for our testing with UMMs. To use this threshold, the
probability for possible test cases need be calculated and compared to this thresh-
old. For example, the probability of the sequence ABCDEBCDC in Fig. 6 can be
calculated as the products of its transitions, that is,
1 × 1 × 0 . 99 × 0 . 7 × 1 × 1 × 0 . 99 × 0 . 3 = 0 . 205821 .
If this is above the overall end-to-end probability threshold, this test case will be
selected and executed.
Coverage, importance and other information or criteria may also be used to gen-
erate test cases. In a sense, we need to generate test cases to reduce the risks
involved in different usage scenarios and product components, and sometimes to
identify such risks as well [14] . The direct risks for selective testing include miss-
ing important areas or not covering them adequately. These “important” areas can
be characterized by various external or internal information. The coverage require-
ment can be handled similarly by adjusting the probabilities to ensure that all
things we would like to cover stays above a certain threshold, or by adjusting
our test case selection procedures. Similar adjustments as above can be used to
ensure that some critical functions of low usage frequencies are also thoroughly
tested.
The hierarchical structure of UMMs also gives us the flexibility to improve test
efficiency by avoiding redundant executions once a subpart has been visited al-
ready. This is particularly true when there are numerous common sub-operations
within or among different end-to-end operations. When revisiting certain states, ex-
act repetition of the execution states that have been visited before is less likely
to reveal new problems. The revisited part can be dynamically expanded to allow
for different lower-level paths or states to be covered. For example, when state
E is revisited in the high-level Markov chain in Fig. 6 , it can be expanded by
using the more detailed Markov chain in Fig. 7 , and possibly execute different sub-
paths there. In general, to avoid exact repetition, we could expand revisited states
with operations of finer granularity, and more thoroughly test those frequently used
parts.
5 . 4 C o n s t r u c t i n g a n d U s i n g U M M s f o r W e b Te s t i n g
Since UMMs are enhanced FSMs, FSM construction described in Section 4 can
be reused as the first step to construct UMMs. The additional step involves assigning
transition probabilities. Similar to the construction for Musa OPs described above,