Abstract

Proof-based System Engineering for Computing Systems and Real-Time Distributed Fault-Tolerant Computing are the areas considered.

The project investigates those algorithmic and methodological issues that arise with mission-critical, complex, computerized applications that may require certification.

Requirements of logical safety, liveness, timeliness and dependability, that are inevitably associated with such applications, can only be met with Real-time Distributed Fault-tolerant computing Technology, hence the ``TRDF'' acronym.

Research work is aimed at breaking new ground in the areas described below.

1) Refinement of a proof-based System Engineering method

State-of-the-art in Computer Science cannot be transferred to users/technology providers unless embedded in a method that can be used by engineers. Furthermore, it is being recognized that the lack of a method for correctly and provably designing and dimensioning mission-critical, complex, computer-based systems is the main reason why a growing number of major failures are being experienced by the industry.

The project has developed a proof-based Systems Engineering method based on models and TRDF algorithms. The TRDF method involves correctness proof obligations. A design correctness proof obligation consists in verifying whether a given design (algorithms + models) solves a given problem. A dimensioning correctness proof obligation consists in verifying whether a valued correct design meets the physical requirements found in the application specification considered.

The specification of a system design/dimensioning that results from applying the TRDF method provably satisfies the specification of the application originally considered.

The TRDF method is being refined in view of potential transfer to external partners. In 1996, the TRDF method has been applied to three real problems, namely Modular Avionics, Safety Systems in Nuclear Power Plants, Safe Landing Gear. It has also been applied to the analysis of the Ariane 5 Flight 501 failure.

2) Composite TRDF algorithms

The goal pursued is to identify, prove and evaluate algorithms and protocols that are solutions to problems arising with real-time, fault-tolerant, distributed/concurrent computations and communications.

Issues of distribution are those arising in the presence of asynchronous parallel computations, with only partial knowledge of global system states. Real-time issues raise the obligation of proving that those timeliness constraints expressed in the specification of some application are always satisfied, for some feasibility conditions.

Fault-tolerance involves demonstrations that correct system behavior is maintained in the presence of given densities of partial failures, for given failure semantics.

For every algorithm/protocol studied, we establish such functions as upper bounds on response times and lower bounds on redundancy. Such functions are established using various techniques (e.g., graph theory, adversary arguments, calculus in (max, +) algebra) and considering deterministic adversaries. We also seek to express distance to optimality (the concept of optimal distributed on-line decision making still is a fundamental research issue). In some instances, we establish that problems have no deterministic solutions.

Examples of results we have established are :

- extensive comparison in terms of complexity (of algorithms, of feasibility condiitons) and efficiency (of algorithms) between fixed - priority scheduling and deadline - driven scheduling,

- upper bounds on response times and feasibility conditions for real-time transactional applications over distributed client-server architectures (Stock Markets, Reservation, Air Traffic Control),

- feasibility conditions for a hybrid off-line/on-line scheduling algorithm aimed at modular avionics systems in the presence of failures.

We also have continued investigating the Asynchronous/Partially Synchronous Group Membership and Consensus problems.

Précédent : Références Remonter : Projet REFLECS, Génie système et