Tactile Wireless Multi-Hop Networks Frank Engelhardt, Mesut Gne - - PowerPoint PPT Presentation

A Latency Analysis of IEEE 802.11-based Tactile Wireless Multi-Hop Networks

Frank Engelhardt, Mesut Güneş Communication and Networked Systems (ComSys) www.comsys.ovgu.de | {fengelha,guenes}@ovgu.de OvGU Magdeburg

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 1

Tactile Wireless Multi-Hop Networks

• Tactile Internet (TI) requires ultra

reliable, ultra low-latency networks

• WMHN are more flexible than

single-hop networks

• 5G will introduce multi-hop

characteristics through Device-2- Device, but question of performance is open

• V2X, Teleoperation, Telesurgery, …

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 2

Tactile Wireless Multi-Hop Networks

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 3

TSE

Master 5G Networking Wireless Multi-Hop Network Latency Slave

Aijaz, A. et al., Realizing the tactile internet: Haptic communications over next generation 5g cellular networks, IEEE Wireless Communications 2017

Single-Flow Latency Model

A haptic flow consists of

• a sub-flow Master→Slave
• another sub-flow Slave→Master

Both sub-flows have the properties

• 1 kHz packet rate
• <100 Bytes per packet (e.g. 6*sizeof(float), a 6 DoF vector)
• requires 1 ms latency bound

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 4 1 R1 H1 Hop: 2 R2 3 h H2 ... R(h-1) ...

Feedback Data Flow Control Data Flow

Master Slave

Single-Flow Latency Model

• End-to-End latency

𝒆𝒇𝟑𝒇 ~ 𝒊 = 𝒍 ∙ 𝒊 → Upper bound for h to reach latency requirement

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 5 1 R1 H1 Hop: 2 R2 3 h H2 ... R(h-1) ...

Feedback Data Flow Control Data Flow

Master Slave

Multi-Flow Latency Model

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 6

• Multiple flows may cross at one or more routers

→𝑒𝑓2𝑓~ ℎ is no longer true

• E.g, a simple example with two flows crossing at router Rx

H1 RA H2 H4 H3 RC RB RD Rx

Multi-Flow Latency Model

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 7

Send

Rx

aSlotTime Receive Send

RD

Receive Send

RB

Receive Send

RC

Receive Send

RA

Receive dmax IFS cw s/r

H3→H4 H3→H4 H1→H2 H1→H2 H2→H1 H2→H1 H4→H3 H4→H3 H3→H4 H3→H4 H1→H2 H1→H2 H2→H1 H2→H1 H4→H3 H4→H3

H1 RA H2 H4 H3 RC RB RD Rx

Multi-Flow Latency Model

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 8

• Multiple flows may cross at one or more routers

→ 𝑒𝑓2𝑓 = 𝑙 ∙ ℎ + 8 ∙ 𝑒𝑄𝑙𝑢,𝑛𝑏𝑦 for this example → 𝑒𝑓2𝑓 = 𝑙 ∙ ℎ + 4 ∙ 𝑜 ∙ 𝑒𝑄𝑙𝑢,𝑛𝑏𝑦 for n intersecting flows

H1 RA H2 H4 H3 RC RB RD Rx

Evaluation of Queueing Delay relative to n

• We model an M/M/1 Queue for the sender of Rx
• Unscheduled WiFi traffic, which has non-deterministic behavior
• Average delay 𝒆𝒃𝒘𝒉 =

𝝁 𝝁(𝝂−𝝁)

with arrival rate 𝝁 and service rate 𝝂

• 𝝂 = 𝒔/𝒕 with transmission bit rate r, packet size s
• 𝝁 = 𝟓 ∙ 𝒐 ∙ 𝝁𝑮𝒎𝒑𝒙, with the packet rate per flow 𝝁𝑮𝒎𝒑𝒙 = 𝟐𝒍𝑰𝒜
• 𝒆𝒃𝒘𝒉(𝒐) =

𝟓∙𝒕∙𝒐∙𝝁𝑮𝒎𝒑𝒙 𝒔𝟑−𝟓∙𝒔∙𝒐∙𝝁𝑮𝒎𝒑𝒙

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 9

Evaluation

• We consider three constellations: Q1, Q2, Q3
• 𝒆𝒃𝒘𝒉(𝒐) =

𝟓∙𝒕∙𝒐∙𝝁𝑮𝒎𝒑𝒙 𝒔𝟑−𝟓∙𝒔∙𝒐∙𝝁𝑮𝒎𝒑𝒙

• Linearization is possible,

e.g. for n<14 for Q3!

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 10

Concluding Remarks

• Linearization could lead to a simple linear model for the entire

network:

𝑒𝐺𝑚𝑝𝑥,𝑏𝑤𝑕 𝑜, ℎ = 𝑙1 ∙ ℎ + 𝑙2 ∙ 𝑜𝐺𝑚𝑝𝑥, with 𝑜𝐺𝑚𝑝𝑥=number of concurrent flows intersecting the current flow

• Model calibration based on simulation and also on real-

hardware testbeds is ongoing work

Frank Engelhardt, Mesut Güneş – 17. GI/ITG KuVS Fachgespräch Sensornetze – FGSN 2018 11