35
Performance Enhancement and Optimized Analysis of the Worst Case End-to-End Delay for AFDX Networks Jian Li, HaiBing Guan, JianGuo Yao , Shanghai Jiao Tong University, Shanghai, China Guchuan Zhu Ecole Polytechnique de Montreal, Canada and Xue Liu McGill University, Canada

Cpscom2012 jianli

  • Upload
    jian-li

  • View
    148

  • Download
    1

Embed Size (px)

DESCRIPTION

This slides is for the presentation in conference CPScom 2012.

Citation preview

Page 1: Cpscom2012 jianli

Performance Enhancement and Optimized Analysis

of the Worst Case End-to-End Delay for AFDXNetworks

Jian Li, HaiBing Guan, JianGuo Yao , Shanghai Jiao Tong University, Shanghai, China

Guchuan ZhuEcole Polytechnique de Montreal, Canada

and Xue LiuMcGill University, Canada

Page 2: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 3: Cpscom2012 jianli

Introduction

• What is AFDX?– Avionics Full Duplex Switched Ethernet (AFDX)

deals with the problems multiplexing huge number of communication flows over a full duplex switched Ethernet network.

Page 4: Cpscom2012 jianli

Introduction

• the context and the entities of AFDX networks– Virtual Link• Data flows in AFDX network• Bandwidth Allocation Gap(BAG)• Maximum Frame Length()

– End System Model– Switch Model

Page 5: Cpscom2012 jianli

Introduction

• the context and the entities of AFDX networks– Virtual Link– End System Model• End System is at the AFDX network boundary in support

of avionics applications accessing the network

– Switch Model

Page 6: Cpscom2012 jianli

Introduction

• the context and the entities of AFDX networks– Virtual Link– End System Model

Page 7: Cpscom2012 jianli

Introduction

• the context and the entities of AFDX networks– Virtual Link– End System Model– Switch Model• The AFDX Switch is responsible for receiving and

forwarding VL packets

Page 8: Cpscom2012 jianli

Introduction

• the context and the entities of AFDX networks– Virtual Link– End System Model– Switch Model

Page 9: Cpscom2012 jianli

Introduction

• End-to-end delay is of a primary concern for AFDX network because of large latency and jitter.– The traditional AFDX traffic model, the leaky-bucket

model(σ,ρ) , is not accurate.– The current switching architecture is not suitable for

various traffic’s priorities with real-time performance requirement in future.

• Development of solution for tightening the estimates of e2e transmission delay in AFDX networks by new model.

Page 10: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 11: Cpscom2012 jianli

AFDX traffic and service model

• Fundamentals of Network Calculus– Definition: Arrival Curve , – Definition: Service Curve , – Theorem 1: Delay Bound ,

Page 12: Cpscom2012 jianli

AFDX traffic and service model

• Network Calculus Representation of AFDX– The arrival curve:• model:• GCRA(Generic Cell Rate Algorithm) model(our new

model) :

– The service curve: )

Page 13: Cpscom2012 jianli

AFDX traffic and service model

• Network Calculus Representation of AFDX– Because of “Pay Burst Only Once” in Network

Calculus. We can get

Page 14: Cpscom2012 jianli

AFDX traffic and service model

• The analysis includes a comparison of the GCRA model and the model

– Delay bound for model:

• The proof is deducted from network calculus 1.4.3

Page 15: Cpscom2012 jianli

AFDX traffic and service model

– Delay bound for GCRA Model

– Proof: the delay bound of the first packet of the flow is

the second packet of the flow, the delay bound is

Then the arrival curve of the GCRA model conforms to the T-SPEC model.So , the delay obtained.■

Page 16: Cpscom2012 jianli

AFDX traffic and service model

– The GCRA model gives to a tighter delay bound than that acquired by the model if the network is heavily loaded.

Page 17: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 18: Cpscom2012 jianli

End-to-end delay with GCRA model

• Single Hop Delay Bound Analysis– The scheduler adopts a Round-Robin policy to

share the transmission capacity among VLs.

Page 19: Cpscom2012 jianli

End-to-end delay with GCRA model

– The service curve offered to a single flow j is – The arrival curve of GCRA model is

– The delay bound is

Page 20: Cpscom2012 jianli

End-to-end delay with GCRA model

• Multi-hop e2e delay analysis

– E2E service curve for a given flow is

– The switch uses a Round-Robin scheduling’s service curve is

Page 21: Cpscom2012 jianli

End-to-end delay with GCRA model

• The e2e delay in GCRA model is

Page 22: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 23: Cpscom2012 jianli

End-to-end delay analysis for FP-RR scheduling

• Single Hop Delay Bound Analysis– FP-RR: all the VLs with the same priority are treated

as an aggregated flow and use Round Robin policy

he

Page 24: Cpscom2012 jianli

End-to-end delay analysis for FP-RR scheduling

– Just like RR scheduling’s analysis , we can get the e2e delay in GCRA model

Page 25: Cpscom2012 jianli

End-to-end delay analysis for FP-RR scheduling

• Multi-Hop E2E Delay Analysis

Page 26: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 27: Cpscom2012 jianli

Simulation

• Platform: TrueTime based on Matlab/Simulink• AFDX network architecture

Page 28: Cpscom2012 jianli

Simulation

• VL configuration

• bandwidth of the network is 100Mb/s

Page 29: Cpscom2012 jianli

Simulation

• The e2e delay with different VL model

Page 30: Cpscom2012 jianli

Simulation

• The e2e delay with FP-RR

Page 31: Cpscom2012 jianli

Simulation

• The e2e delay with FP-RR(ES) and RR(Switch)

Page 32: Cpscom2012 jianli

Simulation

• VL2 and VL10 delay with FP-RR(ES) and R-R(Switch)

Page 33: Cpscom2012 jianli

Agenda

• Introduction• AFDX traffic and service model • End-to-end delay with GCRA model• End-to-end delay analysis for FP-RR scheduling• Simulation• Conclusion

Page 34: Cpscom2012 jianli

Conclusion

• presents a new method for estimating the e2e delay in AFDX networks based on Network Calculus.

• Evaluate the worst case e2e delay when Round-Robin is applied. And show a tighter upper bound in GCRA model.

• development of a new scheduling policy (FP-RR) for ESs and Switches which can get a better e2e delay.

Page 35: Cpscom2012 jianli

Thank you