Worst-case delay analysis of real-time switched Ethernet networks with flow local synchronization

摘要

Full-duplex switched Ethernet is a promising candidate for interconnecting real-time industrial applications. But due to IEEE 802.1d indeterminism, the worst-case delay analysis of critical flows supported by such a network is still an open problem. Several methods have been proposed for upper-bounding communication delays on a real-time switched Ethernet network, assuming that the incoming traffic can be upper bounded. The main problem remaining is to assess the tightness, i.e. the pessimism, of the method calculating this upper bound on the communication delay. These methods consider that all flows transmitted over the network are independent. This is true for flows emitted by different source nodes since, in general, there is no global clock synchronizing them. But the flows emitted by the same source node are local synchronized. Such an assumption helps to build a more precise flow model that eliminates some impossible communication scenarios which lead to a pessimistic delay upper bounds. The core of this thesis is to study how local periodic flows synchronized with offsets can be handled when computing delay upper-bounds on a real-time switched Ethernet. In a first step, the impact of these offsets on the delay upper-bound computation is illustrated. Then, the integration of offsets in the Network Calculus and the Trajectory approaches is introduced. Therefore, a modified Network Calculus approach and a modified Trajectory approach are developed whose performances are compared on an Avionics Full-DupleX switched Ethernet (AFDX) industrial configuration with one thousand of flows. It has been shown that, in the context of this AFDX configuration, the Trajectory approach leads to slightly tighter end-to-end delay upper bounds than the ones of the Network Calculus approach. But offsets of local flows have to be chosen. Different offset assignment algorithms are then investigated on the AFDX industrial configuration. A near-optimal assignment can be exhibited. Next, a pessimism analysis of the computed upper-bounds is proposed. This analysis is based on the Trajectory approach (made optimistic) which computes an under-estimation of the worst-case delay. The difference between the upper-bound (computed by a given method) and the under-estimation of the worst-case delay gives an upper-bound of the pessimism of the method. This analysis gives interesting comparison results on the Network Calculus and the Trajectory approaches pessimism. The last part of the thesis, deals with a real-time heterogeneous network architecture where CAN buses are interconnected through a switched Ethernet backbone using dedicated bridges. Two approaches, the component-based approach and the Trajectory approach, are developed to conduct a worst-case delay analysis for such a network. Clearly, the ability to compute end-to-end delays upper-bounds in the context of heterogeneous network architecture is promising for industrial domains.