Anne-Marie Déplanche

A genetic algorithm for scheduling tasks in a real-time distributed system

Real time systems must often handle several independent periodic macro tasks, each one represented by a general task graph, including communications and precedence constraints. The implementation of such applications on a distributed system communicating via a bus, requires task assignment and scheduling as well as the taking into account of the communication delays. As periodicity implies macro task deadlines, the problem of finding a feasible schedule is critical. The paper addresses this NP hard problem resolution, by using a genetic algorithm under offline and non preemptive scheduling assumptions. The algorithm performance is evaluated on a large simulation set, and compared to classical list based algorithms, a simulated annealing algorithm and a specific clustering algorithm.

STORM a simulation tool for real-time multiprocessor scheduling evaluation

The increasing complexity of the hardware multiprocessor architectures as well as of the real-time applications they support makes very difficult even impossible to apply the theoretical real-time multiprocessor scheduling results currently available. Thus, so as to be able to evaluate and compare real-time multiprocessor scheduling strategies on their schedulability performance as well as energy efficiency, we have preferred a simulation approach and are developing an open and flexible multiprocessor scheduling simulation and evaluation platform called STORM. This paper presents the simulator on which STORM relies and that is able to simulate accurately the behaviour of those (hardware and software) elements that act upon the performances of such systems.

A Study of the AADL Mode Change Protocol

This paper describes a contribution to the verification of AADL models. It focuses on the part of the language dealing with operating modes. An analysis of the AADL mode change protocol is provided. Then, a translation process is exposed, that takes as an input an AADL model and produces as an output a time Petri net. Lastly, it is explained how the resulting time Petri net model can be used to (formally) verify some real-time properties of the AADL model.

Adequacy between AUTOSAR OS specification and real-time scheduling theory

AUTOSAR (AUTOmotive Open System ARchitecture) consortium is a development partnership between the main actors of the automotive manufacturing industry. It aims at defining an open standardized software architecture, in order to face the future challenges in automotive development. One of the important challenge concerns the development of time-critical systems, e.g. brake-by-wire or steer-by-wire. In order to master the development of such systems, one must be able to understand and analyze their real-time behavior. Responses to this problem can be found in the real-time scheduling theory, especially schedulability analysis techniques. In this paper, we propose a review of a subset of the AUTOSAR Operating System specification from a schedulability analysis point-of-view.

Heuristic techniques for allocating and scheduling communicating periodic tasks in distributed real-time systems

This paper deals with the problem of pre-runtime allocating and scheduling communicating periodic tasks in a distributed real-time system. The task system is modeled with independent periodic macro-tasks. The physical architecture consists of a network of identical monoprocessor sites, fully connected by a bus. Two medium access control protocols are considered : CSMA/CA and TDMA. Our objective is to find an allocation of tasks to sites and a subsequent schedule for them such that the periodicity and precedence constraints are satisfied. Besides dealing with these constraints (which is often the sole concern of many studies), the effective communication delays due to the message scheduling and the bus access control protocol are taken into account when the task schedule is being built. Two algorithms for solving this problem are presented: a clustering algorithm and a genetic algorithm.

Implementing a semi-active replication strategy in CHORUS/ClassiX, a distributed real-time executive

The paper reports a practical implementation of a strategy to support semi-active replication of real-time software components (i.e. sets of tasks) running on the Chorus/ClassiX distributed operating system. The main property of the replication strategy developed in this paper is to solve the major difficulty of replica determinism. The semi-active replication scheme consists of a leader software component and identical follower replicas. Only the leader component sends out application messages as well as notifications indicating the order the messages have been consumed and produced. Dynamic non-deterministic scheduling of tasks within the different replicas may cause the follower tasks to lag in their execution regarding the leader ones.

An ADL Centric Approach for the Formal Design of Real-Time Systems

This paper presents the REACT project, dedicated to real-time system design. REACT aims at combining into an architectural design process some formal modelling and verification techniques and providing those corresponding tools. It emphasizes on the ADL of REACT (CLARA), and the validation of functional architectures using formal techniques.

The SCEPTRE2 Communication Service In A Fault Tolerance Context

SCEPTRE2 is a normative report on the basic mechanisms of the real-time executive kernels. No fault tolerance technique exists in its current version. So, in order to develop fault tolerant real time applications, redundancy management services must be provided. In this paper we present some aspects of a fault tolerance implementation in SCEPTRE2 and we focuse particularly on the communication problem, when redundancy exists. Our approach to fault tolerance is based upon the replication of capsules on distinct sites. The adopted replication strategy is the semi-active redundancy. The underlying communication system is synchronous, dependable and implements an atomic broadcastlmulticast. Communication services between capsules ensures that, on one hand, all the capsule replicas have the same inputs, which are used in the same order, and on the other hand sending messages are not duplicated.

Efficiency evaluation of overhead control heuristics in DP-Fair multiprocessor scheduling

A number of optimal algorithms exist for scheduling of periodic taskset with implicit deadlines in real-time multiprocessor systems. However, the practical facts reveal that the optimality is achieved at the cost of excessive scheduling points, migrations and preemptions. In [19], we proposed two heuristics to control the overhead for a class of non-work conserving global scheduling algorithms that combine fluid scheduling and deadline partitioning, while guaranteeing optimality. This paper gives some detailed simulation results along with description of the system to generate the data for the simulation. The given results show the basic strength of the heuristics and validate their efficiency.

Overhead Control Heuristics in Boundary Fair Real-Time Multiprocessor Scheduling

The optimal scheduling algorithms in real-time multiprocessor systems are considered impractical. This is mainly because of the overhead generated due to the frequent scheduling points, migrations and preemptions. The solution to this problem is either to propose new algorithms with less overhead or to improve the existing ones. In this article, some simple heuristics to control the overhead are proposed for a class of optimal algorithms known as Boundary Fair scheduling. The heuristics do not disturb the optimality of the original algorithms. This paper gives some detailed simulation results along with description of experimental conditions. The given results show the basic strength of the heuristics and validate their efficiency both for non-work-conserving and work-conserving systems.