23.03.2017 Views

wilamowski-b-m-irwin-j-d-industrial-communication-systems-2011

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Embedded Networks in Civilian Aircraft Avionics Systems 24-13<br />

The impact of this redundancy mechanism on guaranteed latencies is limited. Of course, average<br />

latency for a single frame improves since Latency = Min (Latency red , Latency blue ), but as we deal with<br />

latency bounds and since latency bounds which are mostly the same on both the red and blue networks,<br />

the final latency bound is unaffected by the redundant network.<br />

24.7 aFDX End-to-End Delay Analysis<br />

As presented before, the main problem is to guarantee that frames exchanged between end <strong>systems</strong> satisfy<br />

the temporal constraints of transported data. Therefore, the end-to-end delay of each path of each<br />

VL has to be studied. The end-to-end delay of each path of VL includes the following characteristics:<br />

• The lower bound for the end-to-end delay corresponds to the case where there is no waiting service<br />

time for the frame in queues. This can be easily computed by adding transmission times over<br />

the physical links and switching delays.<br />

• The upper bound for the end-to-end delay corresponds to the longest aggregate waiting service<br />

time for the frame in queues. This upper bound is mandatory in avionics context for certification<br />

reasons.<br />

Different approaches have been proposed in order to analyze networks flow’s end-to-end delay and<br />

jitter [ACOR99,C98,CSC01,CSEF06,IETF89,IETF98,JNTW02,SCP95,TC94].<br />

Deterministic network calculus (or worst-case network calculus) has been used, for certification reasons,<br />

to compute AFDX upper bounds [C91,C95]. Network calculus theory uses envelopes of arrival<br />

curves and computes a worst-case scenario on each node visited by a flow, taking into account maximum<br />

possible jitter introduced by previously visited nodes [LB98,LBT01]. This approach is obviously<br />

pessimistic as it can lead to impossible scenarios, but it has been useful to prove the determinism of<br />

an AFDX network as it computes a guaranteed upper bound of end-to-end transmission delay for each<br />

VL on the network. This worst-case <strong>communication</strong> delay analysis also gives intermediate information<br />

on latency time in switch output ports, which permits the scaling of the switch memory buffers and<br />

avoiding buffer overflows (i.e., frame loss). Moreover, such an approach has been used for the optimization<br />

of a given AFDX configuration [FFG06]. In conclusion, a certifiable network (with no switch buffer<br />

overflow and bounded latency configuration) often underuses AFDX network capabilities.<br />

Nevertheless, the pessimism of the network calculus approach cannot be evaluated by simply comparing<br />

the end-to-end delay measured on a real AFDX configuration with the computed upper bound<br />

because of the fact that rare events are difficult to observe on a real configuration in a reasonable time.<br />

In order to better understand the real behavior of an AFDX network, a simulation approach has also<br />

been proposed. A distribution of observed end-to-end delay on a simulated AFDX network can thereby<br />

be obtained for each flow [CSF06]. Stochastic network calculus as proposed in [H63,VB01,VB02] might<br />

also help to obtain a probabilistic end-to-end delay analysis [SRF09].<br />

The model-checking approach might help to determine an exact worst-case end-to-end delay and the<br />

corresponding scenario since it explores all possible states of the network. Yet up to now model checking<br />

has not been able to be used on a real-size <strong>industrial</strong> network configuration [CSEF06]. In order to<br />

better evaluate the pessimism of the end-to-end delay upper bound obtained by deterministic network<br />

calculus, the recent trajectory approach seems to be a promising method [MM06]. This approach is<br />

based on the analysis of the worst-case scenario experienced by a packet on its trajectory and not on any<br />

visited node.<br />

24.8 Conclusion<br />

The traditional mono-emitter ARINC 429 data bus has played a major role in the architecture of classic<br />

civilian avionics, but it can no longer cope with the <strong>communication</strong> needs of new avionics architectures<br />

based on IMA. A first solution was to use a multiplexed data bus such as the ARINC 629 data bus,<br />

© <strong>2011</strong> by Taylor and Francis Group, LLC

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!