at Institut National des Sciences Appliquées
Vérification de Systèmes Temporisés Critiques - VERTICS
Expertise and Interests
Since September 2007, I have been an associate professor at the Institut National des Sciences Appliquées(INSA) of Toulouse and I pursued my research at the Laboratoire d’Analyse et d’Architecture des Systèmes (LAAS-CNRS) in the team Vérification de Systèmes Temporisés Critiques (VERTICS). My research deals with:
All my work is related with industrial use-cases, especially in the avionic and automotive fields.
These last years, I have supervised many Mater’s students and two PhD students. The first one, Ahmad Al Sheikh, defended in 2011 and worked on resource allocation in hard real-time avionic systems. The second was addressing the performance evaluation of real-time multiprocessor schedulers and the PhD work was defended by Maxime Che ́ramy in 2014. These two PhDs was funded by the ”Agence Nationale de la Recherche” (ANR).
The next sections present my research from different points of view to show the variety of the problems and methods considered.
- the usage of operational research in the design of software embedded systems under temporal constraints
- the methods to validate et verify real-time architectures
Easing the configuration of real-time systems with operational research methods
A first contribution of my research addresses the automatic configuration of real-time systems. This aims to automatize the computation of the configuration parameters (i.e. allocation of the tasks, bandwidth of the network, priorities, etc.) of embedded systems under constraints (i.e. occupation rate, deadline of the tasks, communication latency, etc.). I tackled these problems through different methods from the Operational Research such as contraint programming, mixed linear programming or game theory.
Bridging the gap between the theory of scheduling and industrial standards
Real-time systems are strongly linked with industrial practices and platforms. That is why academic methods had to be confronted with industrial use-cases and, conversely, industrial practices had to be formalized to use academic methods. This work can be conducted through the study and usage of standards. For most of my work, I have used the avionic standard ARINC and the automotive standard AUTOSAR to constrain design problems and also show some inconsistencies in these standards.
Studing various schedulability analysis
An important issue in real-time is to prove that a system is schedulable. For this, three main approaches are commonly used to analyze a system: model-checking, simulation, and analytical approaches. Through my work, I address these different approaches by proposing new methods: an extension of an analytical method based on worst-case response time for AUTOSAR systems, the usage of the language Pola to model the scheduling of a system was a deep introduction to model-checking, and during the PhD work of Maxime Che ́ramy a simulator, named SimSo, was developed to evaluate multiprocessor scheduling
Introducing uncertainty in real-time scheduling
Real-time system design and the validation uses pessimistic approaches. These analyses are useful when the need of certification is strong mainly for avionic systems, but it seems unnecessary for soft systems. Indeed, these analyses introduce an underutilization of hardware resources and useless constraints during the design phase. The probabilistic scheduling analyze is not predominant in my work, but I maintain a background activity on this subject. For instance, I studied the impact of sampling in probabilistic analysis and more recently I have used stochastic models of the cache memory consumption to evaluate performance of multiprocessor scheduling algorithms.