Le laboratoire Samovar organise un séminaire transverse
lundi 29 mai de 14h à 15h en G09
présenté par Vincent Leroy (Université de Grenoble).
Title: Debugging applications: a data mining approach
Abstract:
Debugging applications can be a tedious task. Traditional test methods focus
on identifying functional bugs in which a function outputs an erroneous value.
However, more subtle bugs can occur in concurrent systems, such as a function
call in a video decoding application occasionally missing its deadline and
causing a visual artifact. In this seminar, I will show how data mining can
help developers better understand these issues.
In the first part of this presentation, I will consider the case of embedded
systems and multimedia applications. The developer first records a trace of
the application, containing both normal activity and misbehaviors. Our
framework, using a combination of clustering and pattern mining, identifies
erroneous activities and finds which system behavior they are correlated to.
The second part of the presentation will focus on model checking. In this
case, the developer can ask a solver to verify a property (such as B is never
executed after A), and the solver usually returns a counter-example trace
containing hundreds of events. Our approach, based on execution graph
analysis, generalizes and abstracts counter-examples to help developers focus
on the key parts of their program that really impact the bug.