Section: Research Program

Management of large or distributed systems

The generic terms of “supervision” or “management” of distributed systems cover problems like control, diagnosis, sensor placement, planning, optimization, (state) estimation, parameter identification, testing, etc. This research axis examines how classical settings for such problems can scale up to large or distributed systems. Our work will be driven by considerations like: how to take advantage of modularity, how to design approximate management algorithms, how to design relevant abstractions to make large systems more tractable, how to deal with models of unknown size, how to design mechanisms to obtain relevant models, etc.

As more specific objectives, let us mention:

• Parametric-size systems: how to verify properties of distributed systems with an unknown number of components;

• Approximate management methods: we will explore the extension of ideas developed for Bayesian inference in large-scale stochastic systems (such as turbo-algorithms) to the field of modular dynamic systems. When component interactions are sparse, even if exact management methods are unaccessible (for diagnosis, planning, control, etc.), good approximations based on local computations may be accessible;

• Model abstraction: we will explore techniques to design more tractable abstractions of stochastic dynamic systems defined on large sets of variables;

• Self-modelling, which consists in managing large-scale systems that are known by their building rules, but where the specific instance is only discovered on-the-fly at runtime. The model of the managed system is built on-line, following the needs of the management algorithms;

• Distributed control: we will tackle issues related to asynchronous communications between local controllers, and to abstraction techniques allowing to address large systems;

• Test and enforcement: we will tackle coverage issues for the test of large systems, and the test and enforcement of properties for timed models, or for systems handling data.