Constraint-Based Testing and Tail Tests for Measurement-Based Probabilistic Timing Analysis

The Worst-Case Execution Time (WCET) of tasks is an important
data to give confidence that Real Time Systems will meet its timing
requirements. Unfortunately, due to its tractability, this data is generally
unknown. Measurement-Based Timing Analysis (MBTA), which
relies on observing execution times driven by test data, has become a
promising approach in the recent years. Some of the current testing
techniques may take a relative long time at triggering decisions because
they require very specific data of the input space. Conversely,
Constraint-Based Testing (CBT) can cope better with these decisions
as well as being more eficient. State-of-the-art approaches integrating
CBT with MBTA have applied code coverage metrics designed
for functional testing but not for WCET. Therefore, important functions
such as generating test data for a path potentially leading to
a large execution time are omitted. A central contribution of this
work embraces CBT. Its objective is to meet code coverage needs
for WCET. The evaluation compares this approach to state-of-theart
Search-Based Testing (SBT) in MBTA and Random Testing (RT)
methods and shows that, in most cases, CBT not only does not underestimate
the largest observed execution time but also it achieves
this data earlier.
A downside of MBTA is that it normally underestimates the WCET.
To face this issue execution time data is recently combined with probabilistic
models. The current probabilistic protocol is hard to automate.
Others probabilistic approaches have found alternative ways to
achieve similar results automatically. The second main contribution
aims for integrating this latter protocol and evaluating its applicability.
The evaluation, which uses execution time data from test generators,
shows that SBT and RT are more likely to generate data that
enable this new approach. The WCET predictions are found more
accurate by definition.