Global State Estimates for Distributed Systems

IEEE Conference on Decision and Control and European Control Conference

Marchand

et al. 2011

Abstract-We consider the control of distributed systems composed of subsystems communicating asynchronously; the aim is to build local controllers that restrict the behavior of a distributed system in order to satisfy a global state avoidance property. We model our distributed systems as communicating finite state machines with reliable unbounded FIFO queues between subsystems. Local controllers can only observe their proper local subsystems and do not observe the queues. To refine their control policy, they can use the FIFO queues to communicate by piggybacking extra information to the messages sent by the subsystems. We define synthesis algorithms allowing to compute the local controllers. We explain how we can ensure the termination of this control algorithm by using abstract interpretation techniques, to overapproximate queue contents by regular languages. An implementation of our algorithms provides an empirical evaluation of our method. I. INTRODUCTIONIn the framework of control of distributed systems, two classes of systems are generally considered, depending on whether the communications between subsystems are synchronous or not. When the synchrony hypothesis [3] can be made, the decentralized control problem and the modular control problem address the design of coordinated controllers that jointly ensure the desired properties for this kind of systems [26], [22], [21], [9], [13]. When considering asynchronous distributed systems, one have to take into account some communication delays between the components of the system, which renders the distributed control problem much harder even undecidable [24].We are here interested in the second problem i.e., the distributed control problem. Our aim is to solve this problem when the system to be controlled is composed of n subsystems that asynchronously communicate through reliable unbounded FIFO channels (or queues). These subsystems are modeled by communicating finite state machines [5] (CFSM for short) that explicitly express the work and communications of a distributed system. This model appears to be essential for concurrent systems in which components cooperate via asynchronous message passing through unbounded buffers (they are e.g. widely used to model communication protocols). We thus assume that the distributed system is already built and the architecture of communication between the different subsystems is fixed. Following the architecture described in Figure 1, we assume that each subsystem is

“…In this section, we informally explain the SE-algorithm that allows each local controller to compute its own state estimate. More details can be found in [12]. Vector Clocks.…”

Section: A State Estimates Algorithmmentioning

confidence: 99%

Section: A State Estimates Algorithmmentioning

confidence: 99%

Synthesis of communicating controllers for distributed systems

Kalyon¹,

IEEE Conference on Decision and Control and European Control Conference

Marchand

et al. 2011

“…This paper is an extended version of two conference papers [18] and [17]. It provides the full process allowing to derive controllers from a state-based specification and a plant by means of state-based estimates and abstract interpretation techniques, whereas [17] was only presenting the state-based algorithms and [18] the control point of view with an overview of the state-based estimates computation point of view. The proofs absent from this paper are available in [16].…”

Section: Introductionmentioning

confidence: 99%

Symbolic Supervisory Control of Distributed Systems With Communications

Kalyon

IEEE Trans. Automat. Contr.

Marchand

et al. 2014

Self Cite

“…If a CM system does not satisfy a safety property, control automatically computes a restriction of this system that assures safety by implementing the distributed control algorithm presented in [7] (see control(1) man page for details). Figure 1 shows the modular architecture of McScM's verify tool.…”

Section: Safety Verification Of Communicating Machinesmentioning

confidence: 99%

“…McScM is programmed in OCaml and available under a BSD license. The development of McScM takes place in a software forge [14] that provides a wiki for documentation (including man pages and API), a bug and issue tracker, as well as our theoretical work [6,7]. The following discussion refers to McScM's release 1.2.…”

Section: Introductionmentioning

confidence: 99%

McScM: A General Framework for the Verification of Communicating Machines

Heußner

Tools and Algorithms for the Construction and Analysis of Systems

Sutre

2012

Self Cite

Abstract. We present McScM, a platform for implementing and comparing verification algorithms for the class of finite-state processes exchanging messages over reliable, unbounded FIFO channels. McScM provides tools for the safety verification and controller synthesis of these infinite-state models. Our verification tool implements several modelchecking techniques: CEGAR with different abstraction-refinement methods, abstract interpretation, abstract regular model checking, and lazy abstraction. Seen as a general framework for the class of transition systems with finite control/infinite data, McScM delivers the basic infrastructure for implementing verification algorithms, and privileges to conveniently implement new ideas on a high level of abstraction. It also allows us to compare and benchmark different algorithmic approaches with the same backend.