|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Diagnosis of large scale discrete event systems
Participants : Marie-Odile Cordier, Christine Largouët, Sophie Robin, Laurence Rozé, Yulong Zhao.
The problem we deal with is monitoring complex and large discrete-event systems (DES) such as an orchestration of web services or a fleet of mobile phones. Two approaches have been studied in our research group. The first one consists in representing the system model as a discrete-event system by an automaton. In this case, the diagnostic task consists in determining the trajectories (a sequence of states and events) compatible with the sequence of observations. From these trajectories, it is then easy to determine (identify and localize) the possible faults. In the second approach, the model consists in a set of predefined characteristic patterns. We use temporal patterns, called chronicles, represented by a set of temporally constrained events. The diagnostic task consists in recognizing these patterns by analyzing the flow of observed events.
Distributed monitoring with chronicles - Interleaving diagnosis and repair - Making web services more adaptive
Our work addresses the problem of maintaining the quality of service (QoS) of an orchestration of Web services (WS), which can be affected by exogenous events (i.e., faults). The main challenge in dealing with this problem is that typically the service where a failure is detected is not the one where a fault has occurred: faults have cascade effects on the whole orchestration of services. We have proposed a novel methodology to treat the problem that is not based on Web service (re)composition, but on an adaptive re-execution of the original orchestration. The re-execution process is driven by an orchestrator Manager that takes advantage of an abstract representation of the whole orchestration and may call a diagnostic module to localize the source of the detected failure. It is in charge of deciding the service activities whose results can be reused and may be skipped, and those that must be re-executed.
This year, we have improved the prototype, adding the visualization of the roadmap and the activities that do not have to be reexecuted. This work has been published in ICWS2013 [15] and we are working on a journal paper that will be submitted in 2014.
Scenario patterns for exploring qualitative ecosystems
This work aims at giving means of exploring complex systems, in our case ecosystems, and more recently agrosystems, specifically herd management systems. We proposed to transform environmental questions about future evolution of ecosystems into formalized queries that can be submitted to a simulation model. The system behavior is represented as a discrete event system described by a set of interacting timed automata, the global model corresponding to their composition on shared events. To query the model, we have defined high-level generic query patterns associated to the most usual types of request scenarios. These patterns are then translated into temporal logic formulas. The answer is computed thanks to model-checking techniques that are efficient for analyzing large-scale systems. Five generic patterns have been defined using TCTL (Timed Computation Tree Logic) “WhichStates”, “WhichDate”, “Stability”, “Always”, “Safety”. Three of them have been implemented using the model-checker UPPAAL.
The approach has first been experimented on a marine ecosystem under fishing pressure. The model describes the trophodynamic interactions between fish trophic groups as well as interactions with the fishery activities and with an environmental context. A paper has been previously published in the Environmental Modelling Software Journal [65] . More recently, a similar approach has been experimented on agrosystems, specifically herd management systems, for which a hybrid model has been built using hierarchical timed automata. This later work has been achieved in the context of Yulong Zhao's PhD thesis [6] and done in collaboration with our colleagues of Inra .
Controler synthesis for dealing with “How to” queries
We extended the approach to deal with “How to” queries. As before, we rely on a qualitative model in the form of timed automata and on model-checking tools to answer queries. We proposed and compared two approaches to answer questions such as “How to avoid a given situation ?”(safety query). The first one exploits controller synthesis and the second one is a “generate and test” approach. We evaluated these two approaches in the context of an application that motivates this work, i.e. the management of a marine ecosystem and the evaluation of fishery management policies. The results have beenpreviously published in [88] .
More recently, we used similar methodological tools to analyze in the context of herd management on a catchment. An hybrid model has been built using hierarchical timed automata and scenarios can be simulated and evaluated using the approach presented in the previous paragraph. In this context, the goal is to identify and analyse the best/optimal farming practices in order to reduce nitrate pollution due to livestock effluents. We proposed to use controler synthesis tools and to couple them with machine learning techniques in order to get the best strategies and to put them on easy-to-use form. This work has been made in the context of Yulong Zhao's PhD thesis [6] and in collaboration with our colleagues of Inra (Umr Pegase).