Frédéric Ridouard

A Probabilistic Analysis of End-To-End Delays on an AFDX Avionic Network

AFDX (Avionics Full DupleX Switched Ethernet, ARINC 664) developed for the Airbus A380 represents a major upgrade in both bandwidth and capability. Its reliance on Ethernet technology helps to lower some implementation costs, but guaranteed service presents challenges for system designers. An analysis of end-to-end transfer delays through the network is required in order to determine upper bounds. In this paper, we propose to compute probabilistic upper bounds for end-to-end delays on avionic flows. Such upper bounds can be exceeded with a given probability p, and are relevant in the context of avionics, where functions are designed to give accurate results even if they miss some frames. The stochastic network calculus approach analytically determines a probabilistic upper bound, whereas the simulation approach gives an experimental upper bound. The former may be used for new certification needs since it assures that the probability of exceeding the computed upper bound is not greater than p. The latter closely approximates actual network behavior and can help to give some idea of the pessimism of the stochastic network calculus upper bound. The two approaches have been developed in the context of an industrial AFDX network configuration.

Negative results for scheduling independent hard real-time tasks with self-suspensions

In most real-time systems, tasks use remote operations that are executed upon dedicated processors. External operations introduce self-suspension delays in the behavior of tasks. This paper presents several negative results concerning scheduling independent hard real-time tasks with self-suspensions. Our main objective is to show that well-known scheduling policies such as fixed-priority or earliest deadline first are not efficient to schedule such task systems. We prove the scheduling problem to be NP-hard in the strong sense, even for synchronous task systems with implicit deadlines. We also show that scheduling anomalies can occur at run-time: reducing the execution requirement or the suspension delay of a task can lead the task system to be infeasible under EDF. Lastly, we present negative results on the worst-case performances of well-known scheduling algorithms (EDFy RM, DM, LLF, SRPTF) to maximize tasks completed by their deadlines.

Probabilistic upper bounds for heterogeneous flows using a static priority queueing on an AFDX network

AFDX (avionics full duplex switched Ethernet, AR-INC 664) developed for the Airbus A380 represents a major upgrade in both bandwidth and capability. Its reliance on Ethernet technology helps to lower some implementation costs, but guaranteed service presents challenges for system designers. An analysis of end-to-end transfer delays through the network is required in order to determine upper bounds. The stochastic network calculus approach analytically determines worst-case probabilistic upper bounds in the context of homogeneous avionics flows without priorities. Such upper bounds can be exceeded with a given probability PUB, and are relevant in the context of avionics, where functions are designed to give accurate results even if they miss some frames. Nowadays, there is a need to handle new classes of traffics with different priorities (voice, video, best-effort, ...) on the same AFDX network with no consequences on existing avionic flows. This paper presents the application of the stochastic network calculus approach in the context of a static priority queueing service discipline and evaluates the influence of the service discipline on analytical probabilistic upper bounds.

Stochastic network calculus for end-to-end delays distribution evaluation on an avionics switched Ethernet

AFDX (avionics full duplex switched Ethernet, AR-INC 664) used for modern aircraft such as Airbus A380 represents a major upgrade in both bandwidth and capability for aircraft data networks. Its reliance on Ethernet technology helps to lower some of the implementation costs, though the requirement for guaranteed service does present challenges to system designers. Thus, the problem is to prove that no frame will be lost by the network (no switch queue will overflow) and to evaluate the end-to-end transfer delay through the network. Several approaches have been proposed for this evaluation. Deterministic network calculus gives a guaranteed upper bound on end-to-end delays, while simulation produces more accurate results on a given set of scenarios. In this paper, we propose a stochastic network calculus approach in order to evaluate the distribution of end-to-end delays. We evaluate the pessimism of the results on some typical AFDX flows, as described by virtual links.

Optimistic problems in the trajectory approach in FIFO context

AFDX (Avionics Full Duplex Switched Ethernet) is a network designed for safety-critical applications in avionics systems. The flow analysis is mandatory for the certification. The end-to-end (ETE) delay upper bound has to be computed and hence guaranteed. Different approaches have been proposed to compute ETE delay upper bound such as the Network Calculus and the trajectory method. Both methods are deemed to introduce some pessimism in the computations. We only focus next on the trajectory approach. This paper shows that the original trajectory approach is flawed and can compute an optimistic ETE. We present a counter-example and discuss the sources of optimism that must be considered to revise the trajectory method.

Improving AFDX end-to-end delays analysis

The AFDX (Avionics Full DupleX Switched Ethernet) is the backbone network of most modern civilian aircraft. For certification reasons, a guaranteed upper bound has to be determined on the end-to-end delay for all transmitted flows. Several approaches have been designed in order to analyze such delays. In this paper, we focus on the Forward End-to-End delays Analysis (FA). This method is based on real-time scheduling results. Our goal is to propose an improvement, that reduces the pessimism of the computed bounds, compared to earlier versions of the method. The improvement relies on the way the serialization of packets sharing a common link is taken into account. We also propose a comparative analysis on a simple configuration with two other methods (Network Calculus and the Trajectory approach).

Job partitioning strategies for multiprocessor scheduling of real-time periodic tasks with restricted migrations

In this work, we consider the scheduling of real-time periodic tasks on multiprocessors under a restricted-migration constraint (tasks are allowed to migrate whereas jobs are not). We first propose a technique, (offline) job partitioning, which statically assigns jobs to processors before execution. We show that (offline) job partitioning strictly dominates both partitioned and restricted-migration scheduling. We then consider r-RM, a restricted-migration variant of global RM, we design a sufficient schedulability test and we provide a speedup factor for the scheduler.

A Forward end-to-end delays Analysis for packet switched networks

Packet switched networks such as AFDX (Avionics Full Duplex Switched Ethernet) are a major upgrade for avionics systems communication. The computation of a guaranteed upper bound on End-to-end delay of all flows transmitted in the network is mandatory for certification reasons. Considering existing methods, we define a forward End-to-End delay analysis that can scale to large configurations and high loads. The proposed approach is compared with Network Calculus and Trajectory approach on an AFDX configuration.

Existing offset assignments are near optimal for an industrial AFDX network

Avionics Full DupleX Switched Ethernet (AFDX) has been developed for modern aircraft such as Airbus 380. Due to the non-determinism of switching mechanism, a worst-case delay analysis of the flows entering the network is a key issue for certification reasons. Up to now most existing approaches (such as Network Calculus) consider that all the flows are asynchronous and they do not take into account the scheduling of flows generated by the same end system. It is then pessimistic to take into account such a synchronous scenario. Each end system can be considered as an offset free system, thus the main objective of this paper is to evaluate existing offset assignments in the context of an industrial AFDX network. Existing offset assignments are adapted to take into account specific characteristics of an AFDX network. Worst-case delay results are obtained according to these offset heuristics. It is shown that some existing heuristics are not efficient while some are near optimal for the studied industrial AFDX network.

Forward End-To-End delay Analysis for AFDX networks

Packet switched networks and message multiplexing have been a major upgrade for industrial systems communications. In the avionics domain, this evolution was brought by the introduction of avionics full duplex switched ethernet (AFDX). Guaranteed upper bounds of end-to-end delays for messages transmitted over an AFDX network are mandatory for certification reasons. In this article, we present the forward end-to-end delay analysis (FA), which is scalable to both large and heavily loaded configurations. A formal proof of FA is detailed and the approach is compared with alternative methods (network calculus and the trajectory approach) on two types of AFDX configurations (including a real industrial architecture).