Tim iming analysis of f AVB Ethernet network usin ing the Forward end-to to-end Dela lay Analysis Nassima BENAMMAR Frédéric Ridouard Henri Bauer Pascal Richard Oc October 12, 12, 2018 2018 1
Context A UTOMOTIVE S YSTEM : ECU S ECU : Electronic Control Unit 2
Context E XPLOSION OF AUTOMOTIVE SYSTEMS 100% autonomous cars for public Complete autonomous capability Limited driver substitution Critical real time systems Passive autonomous driving 2023 2012 2018 2019 2016 2017 http://www.caam.org.cn/english/ 3
Context D ISTRIBUTED ARCHITECTURE 4
Context R EAL TIME NETWORKS IN D ISTRIBUTED ARCHITECTURE Deterministic ; Security, fault tolerance ; Bandwidth; Easy integration; Decrease wiring. 5
Context A UTOMOTIVE NETWORKING Heterogeneous Complexity networking Trends to Ethernet based standards Audio Video Bridging Switched Ethernet (AVB) Time Sensitive Networking (TSN) 6
Context A UDIO V IDEO B RIDGING S WITCHED E THERNET (AVB) • Synchronization ; • Quality of service ; • SR : Stream reservation. SR Class A Shaper Based Credit (CBS) FPFIFO SR Class B Best Effort 7
Problematic R EAL TIME SYSTEMS ANALYSIS End-to-end Real time Traffic shaper delay constraints 8
Problematic T IMING ANALYSIS Maximum jitter Mo Model Tim iming anal analysis base based on on checkin ing mathematics Distribution of end-to-end delay Worst case delay Exact worst Minimum delay Maximum delay by computed by timing case delay simulation analysis 9 Deadline
Motivations AVB FIFO CBS FP/FIFO NC • Network Calculus (NC) • Trajectory approach (TA) TA • Compositionnal Performance Analysis CPA (CPA) • Forward end-to-end delay Analysis (FA) FA Existing Our contribution 10
Summary ry - FA METHOD - FA EXTENSION TO CBS ALGORITHM - E XPERIMENTATIONS - C ONCLUSION 11
FA method 12
FA method FA extension to CBS algorithm Experimentation Conclusion N ETWORK MODEL OF FA node 3 node 1 node 2 node 4 Node = multiplexing point 13
FA method FA extension to CBS algorithm Experimentation Conclusion E ND - TO - END DELAY ELEMENTS Source node . . . . . . . . . . . . L L Variable delay Variable delay Variable delay L: technological latency 14
FA method FA extension to CBS algorithm Experimentation Conclusion P ROPERTIES A flow 𝑤 𝑗 is defined by : ℎ : maximum transmission time of a frame from 𝑤 𝑗 in the node ℎ . • 𝐷 𝑗 • 𝑈 𝑗 : period of 𝑤 𝑗 . • 𝑄𝑏𝑢ℎ 𝑗 : path of 𝑤 𝑗 . • 𝑄 𝑗 : priority level of 𝑤 𝑗 . priority 1 ℎ : set of flows 𝑤 𝑘 ∈ Γ ℎ où 𝑄 𝑗 = 𝑄 • 𝑡𝑞 𝑗 𝑘 . ℎ : set of flows 𝑤 𝑘 ∈ Γ ℎ où 𝑄 𝑗 < 𝑄 • ℎ𝑞 𝑗 𝑘 . FPFIFO priority 2 ℎ : set of flows 𝑤 𝑘 ∈ Γ ℎ où 𝑄 𝑗 > 𝑄 • 𝑚𝑞 𝑗 𝑘 . …. priority n Fixed Priority First In First Out (FPFIFO) 15
FA method FA extension to CBS algorithm Experimentation Conclusion A NALYSIS OF A FRAME WITH FA Generation instant time of 𝑔 𝑗 𝑔𝑗𝑠𝑡𝑢 𝑗 ℎ 𝑢 𝑋 𝑗 Worst case arrival instant 𝑔 𝑗 ℎ L Worst case waiting time ℎ + 1 Worst case arrival instant ℎ 𝑢 = 𝑋𝑀𝑄 𝑗 ℎ 𝑢 + 𝑋𝑇𝑄 𝑗 ℎ 𝑢 + 𝑋𝐼𝑄 𝑗 ℎ 𝑢 𝑋 𝑗 L: technological latency 16
FA method FA extension to CBS algorithm Experimentation Conclusion W ORST CASE WAITING TIME FOR A FRAME IN A NODE DURING A TIME INTERVAL Assumption : one level of priority (FIFO) ℎ 𝑢 : transmission time of all frames generated from 𝑤 𝑘 in the node ℎ during 𝑢 . • 𝑠𝑐𝑔 𝑘 𝑔 𝑗 𝑔 ℎ 𝑗 𝑢 0 𝑋 ℎ 𝑢 − 𝑢 − 𝐷 𝑗 ℎ 𝑋 ℎ 𝑢 = ℎ (𝑢) 𝑠𝑐𝑔 𝑘 𝑤 𝑘 ∈Γ ℎ 17
FA method FA extension to CBS algorithm Experimentation Conclusion M AXIMUM INTERFERENCE GENERATED IN A NODE DURING 0, 𝑢 ℎ during 𝑢 . Given a frame 𝑔 𝑗 generated from 𝑤 𝑗 in the node ℎ incoming from an imput port 𝐽𝑄 1 Burst of frames ℎ 𝑢 = 𝑋𝑀𝑄 𝑗 ℎ 𝑢 + 𝑋𝑇𝑄 𝑗 ℎ 𝑢 + 𝑋𝐼𝑄 𝑗 ℎ 𝑢 𝑋 𝑗 …. ℎ 𝑢 𝑔 𝑗 ℎ − 1 𝑢 𝑔 L 𝑗 L 𝑔 𝑗 ℎ 𝑢 0 18 18
FA method FA extension to CBS algorithm Experimentation Conclusion W ORST CASE WAITING TIME IN A NODE ℎ 0 𝑢 𝑋 ℎ (𝑢) ℎ 𝑢 − 𝑢) ℎ 𝑢 ≥ t 𝑛𝑏𝑦 𝑢≥0 (𝑋 Maximum delay incurred in the node ℎ = W hile 𝑋 𝑗 𝑗 19
FA extension to CBS algorithm 20
FA method FA extension to CBS algorithm Experimentation Conclusion C BS A LGORITHM SRA SRA BE SRB SRA 𝑢 0 Credit A Credit B 0 𝑢 BE SRA SRA SRB SRA 𝑢 𝑢 2 𝑢 4 𝑢 5 0 𝑢 1 𝑢 3 A flow 𝑤 𝑗 belongs to a SR class 𝑌 𝑗 ℎ 𝑢 = 𝑋𝑀𝑄 𝑗 ℎ 𝑢 + 𝑋𝑇𝑄 𝑗 ℎ 𝑢 + 𝑋𝐼𝑄 𝑗 ℎ 𝑢 +𝑋𝑂𝐽𝑀 𝑗 ℎ 𝑢 𝑋 21 𝑗
FA method FA extension to CBS algorithm Experimentation Conclusion I MPACT OF SAME PRIORITY FLOWS 𝑜 𝐷𝑠𝑓𝑒𝑗𝑢 𝑌 𝑗 ℎ 𝑢 = ℎ 𝑢 , 𝑀𝑗𝑜𝑙𝑆𝑏𝑢𝑓 𝑗,𝑦 ℎ 𝑢 , 𝑇ℎ𝑏𝑞𝑓𝑠𝑆𝑏𝑢𝑓 𝑗,𝑦 ℎ (𝑢)) 𝑋𝑇𝑄 𝑗 min(𝑠𝑐𝑔 𝑗,𝑦 ℎ 𝐷𝑠𝑒 𝑌 𝑗 ,𝑛𝑏𝑦 𝑦=1 ℎ,+ 𝐷𝑠𝑓𝑒𝑗𝑢 𝑌 𝑗 𝜏 𝑌 𝑗 Busy period of ℎ 𝐷𝑠𝑒 𝑌 𝑗 ,𝑛𝑏𝑦 𝑛𝑏𝑦 𝑤 𝑘 ∈𝑀𝑄 𝑌𝑗 ℎ 𝐷 𝑘 ℎ,− ℎ 𝐼𝑄 - 𝜏 𝑌 𝑗 𝑌 𝑗 𝑛𝑏𝑦 𝑤 𝑘 ∈𝑌 𝑗 𝐷 𝑘 0 𝑢 (i) (iii) (ii) 22
FA method FA extension to CBS algorithm Experimentation Conclusion I MPACT OF HIGH PRIORITY FLOWS Given a frame 𝑔 𝑗 generated from 𝑤 𝑗 belonging to an SR Class 𝑌 𝑗 in the node ℎ during 𝑢 . ℎ 𝑢 = 𝑛𝑗𝑜( ℎ 𝑢 ℎ 𝑢 + 𝑛𝑏𝑦 𝑤 𝑘 ∈𝑇𝐼𝑄 𝑍 ℎ ℎ (𝑒𝑓𝑞 𝑗 𝑋𝐼𝑄 𝑠𝑐𝑔 𝑒𝑓𝑞 𝑗 , 𝐶𝑝𝑠𝑒 𝑍 ℎ 𝐷 𝑘 )) 𝑗 𝑘 ℎ 𝐷𝑠𝑓𝑒𝑗𝑢 𝑌 𝑗 ℎ 𝑍∈𝐼𝑄 𝑌𝑗 𝑤 𝑘 ∈ℎ𝑞 𝑗 ℎ 𝐷𝑠𝑒 𝑌 𝑗 ,𝑛𝑏𝑦 ℎ,− - 𝜏 𝑌 𝑗 Burst of frames 𝑛𝑏𝑦 𝑤 𝑘 ∈𝑌 𝑗 𝐷 …. 𝑘 0 𝑢 ℎ 𝑔 (i) (iii) (ii) 𝑗 𝑢 𝑔 𝑗 ℎ ℎ (𝑢) 𝑢 0 𝑋 Non serialized part Serialized part 𝑗 23 ℎ (𝑢) 𝑒𝑓𝑞 𝑗
FA method FA extension to CBS algorithm Experimentation Conclusion T AKING INTO ACCOUNT RESPLENISHING TIME OF THE CREDIT OF 𝑌 𝑗 ℎ 𝑢 𝑋𝐼𝑄 𝑗 𝐷𝑠𝑓𝑒𝑗𝑢 𝑌 𝑗 𝜏 − 𝜏 − 𝐷 𝑙 𝐷 𝜏 + 𝑘 𝐷 𝑙 𝜏 + 𝐷 𝑘 𝑔 𝑔 𝑘 𝑙 0 𝑢 - 𝜏 − 𝜏 + 1 + 𝜏 − ℎ 𝑢 ℎ 𝑢 + 𝑋𝑂𝐽𝑀 𝑗 ℎ 𝑢 𝑋𝑇𝑄 𝑗 𝑋𝑇𝑄 𝑗 𝜏 + 24
Experimentations 25
FA method FA extension to CBS algorithm Experimentations Conclusion I NDUSTRIAL CASE 𝒘 𝒋 𝑼 𝒋 (𝝂𝒕) 𝑫 𝒋 (𝝂𝒕) 𝑺 𝒋 (𝝂𝒕) Class Source Destination MM_Disk 𝒘 𝟐 RSE 5000 5,12 A 2762,72 Gateway CPU Gateway Head_Unit 𝒘 𝟑 5000 5,12 A 2994 Gateway RSE 𝒘 𝟒 125 32,64 A 188,96 DA_CAM Heand_Unit DA_CAM 𝒘 𝟓 280 115,68 B 631,9 MM_Disk RSE Amplifier 𝒘 𝟔 TV 1400 115,68 B 618,67 MM_Disk Amplifier 𝒘 𝟕 560 115,68 B 395,68 TV Head_Unit CPU Technological latency= 8𝜈𝑡 . • • [Li et al., Henia et al. , Steinbach et al. ] Service rate= 100 Mbit/s. 26
FA method FA extension to CBS algorithm Experimentations Conclusion C OMPARISON WITH CPA Technological latency = 0,33𝜈𝑡 . • • Service rate = 1 Gbit/s. I/O I/O I/O I/O I/O I/O 𝑇𝑥𝑗𝑢𝑑ℎ 1 𝑇𝑥𝑗𝑢𝑑ℎ 16 𝑇𝑥𝑗𝑢𝑑ℎ 1 𝑇𝑥𝑗𝑢𝑑ℎ 4 Ctrl Ctrl I/O I/O I/O I/O I/O I/O I/O I/O Linear topology 27 Clustered topology [Diemer et al. ]
FA method FA extension to CBS algorithm Experimentations Conclusion R ESULTS Worst case delay( 𝜈𝑡 ) Flow 28 Linear topology
Conclusion 29
FA method FA extension to CBS algorithm Experimentation Conclusion C ONCLUSION • Tight bounds when taking into account : • the serialization effect ; • the shaper on the maximum interference from high and same priority flows ; • and the resplenishing time of the credit without counting twice the interference of high priority flows. • Interesting results compared to CPA. 30
FA method FA extension to CBS algorithm Experimentation Conclusion P ERSPECTIVES • Extension of FA method to the servicing policy of TSN standard (Time Aware Shaper). TDMA Critical traffic Time aware shaper Classe SR A Shaper Based Credit (CBS) Classe SR B Flux BE 31
Thank you for your attention Questions? N ASSIMA . BENAMMAR @ ENSMA . FR 32
C OMPLEXITY OF AUTOMOTIVE SYSTEMS : DISTRIBUTED ARCHITECTURE 33 https://www.st.com/en/solutions-reference-designs/automotive-solutions.html
Recommend
More recommend
