winlab 1 roy yates ece/winlab, rutgers nsf workshop on ultra low latency wireless networks march 26,...

34
WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and Networks

Upload: ami-carson

Post on 13-Jan-2016

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

1

Roy YatesECE/WINLAB, Rutgers

NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015

Age of Information Status Updating Systems and Networks

Page 2: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

2

Motivation

• 50 years of rate maximization– at the expense of delay

• long (coded) packets on wireless channels,• ARQ • video streaming with large delays to absorb

packet jitter• Caching to compensate for network latency

• high throughput “best-effort” networks

Page 3: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

3

Applications(The Samsung 5G List)

• V2X – timely delivery of critical messages for traffic safety

• Mission-Critical IoT (M2M)– mission-critical systems – process monitoring/detection and

disaster response

• Virtual/Augmented Reality– seamless virtual/real-world interaction

• Real-time remote access (tactile feedback)– long range, real-time control for remote surgery, driving,

etc.

• Everything-on-Cloud– instantaneous cloud-based services/multimedia content

~1 ms

1-10 ms

10-100ms

Page 4: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

4

Applications(The Samsung 5G List)

• V2X – timely delivery of critical messages for traffic safety

• Mission-Critical IoT (M2M)– mission-critical systems – process monitoring/detection and

disaster response

• Virtual/Augmented Reality– seamless virtual/real-world interaction

• Real-time remote access (tactile feedback)– long range, real-time control for remote surgery, driving,

etc.

• Everything-on-Cloud– instantaneous cloud-based services/multimedia content

~1 ms

1-10 ms

10-100ms

57 mph=

1 inch/ms

Page 5: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

5

Applications(The Samsung 5G List)

• V2X – timely delivery of critical messages for traffic safety

• Mission-Critical IoT (M2M)– mission-critical systems – process monitoring/detection and

disaster response

• Virtual/Augmented Reality– seamless virtual/real-world interaction

• Real-time remote access (tactile feedback)– long range, real-time control for remote surgery, driving,

etc.

• Everything-on-Cloud– instantaneous cloud-based services/multimedia content

• Wireless Network on Chip – Cloud on Chip?

~1 ms

1-10 ms

10-100ms

?? ms

Page 6: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

6

Network Delay(H. Viswanathan, Bell Labs)

Page 7: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

7

Remote Surgery

Page 8: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

8

Remote Surgery

Page 9: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

9

PHY

• Wider Channels (but not UWB )– 30+ GHz mmWave

• M2M: reliability is essential

• Practice: Emerging low latency 5G– Channel Estimation, Modulation, Coding,

Framing • (check out Fettweis CTW 2013 talk)

• Theory:– Delay-Limited Capacity, Short blocklength

source/channel coding

Page 10: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

10

Latency-Sensitive MAC

• Practice:– 2G/3G packet voice MAC– LTE Scheduling– Sleep protocols!

Page 11: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

11

Latency-Sensitive MAC

• Practice:– 2G/3G packet voice MAC– LTE Scheduling– Sleep protocols!

• What Randy said: “CSMA style random access seems ill-matched to low latency unless the network is very underutilized.”

• Theory: Are rate/delay tradeoffs fundamental?

Page 12: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

12

Large Networks (Hundreds of cars)

Frequent Updates

Reliability and Timeliness are required

V2V Safety Messaging

Source

Page 13: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

13

• DSRC standard MAC protocol• Message Scheduling, Forwarding/Piggybacking• Power/rate adaptation, coverage …

• Performance Metrics?

V2V Safety Messaging

Source

Page 14: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

14

• Car u sends updates to car v• Updates pass through network/service

system• Car v wants latest state information.

• Metric: Age of the latest status update

Network

Page 15: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

15

Update Age

D(t) UpdateSent

Received

tt1 t2t1’ t2'

Page 16: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

16

Update Age

D(t)

• Low Update Rate Age gets large between

updates

UpdateArrival

Departure

tt1 t2t1’ t2'

Page 17: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

17

Update Age

D(t)

• High Update Rate Queueing Delay

t1 t2 t1’ t2't3 t3'

Page 18: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

18

Average Update Age

D(t)

• Update Rate:• High Queueing delays• Low Infrequent

updates

High Average Age

Average Age

Page 19: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

19

FCFS Average Update Age

D(t)

𝑋 𝑇

• X= Interarrival Time

• T= System Time

• Weak ergodicity requirements• E[XT ] is tricky, negative correlation

Page 20: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

20

Average Age

Page 21: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

21

Average Age

Nothroughput/

delaytradeoff

Page 22: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

22

Other Age Metrics

Average Peak AgeD1

D(t)

D(t)

D2

D3

D* P[ (D t)>D* ]

Page 23: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

23

Competing Updates

• How often is too often?

Page 24: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

24

Multiple Sources

Page 25: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

25

Multiple Sources

Models for Source 2:• Competing status updater• Other traffic

Page 26: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

26

Multiple SourcesFCFS Status Age Region

OptimalSharing

Nash Equilibriu

m

Page 27: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

27

𝜇Source

1

Source2

Monitor

• Queueing delays increases status age

• Reduce/Eliminate the queues?• “Packet Management”

[Costa, Codreanu, Ephremides ISIT’14]

Page 28: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

28

LCFSPre-emption & Discarding

(No Queueing)

𝜇Source

1

Source2

Monitor

l1

l2

Page 29: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

29

V2V Safety Messaging

Network

• Multiple Sources• Fast local server interface • Slow Server

Page 30: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

30

Multiple SourcesFCFS/LCFS Age

Page 31: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

31

Timely CompressionA Status Updating Problem

• Encoder input symbol = status update • Age = Decoder symbol lag• Block coding Bursty bit arrivals at FIFO buffer

Bit pipe queueing delay Decoding delay

• [Sahai&Cheng ISIT’07]

a1a2a3…

Encoder FIFObuffer

Rate Rbit pipe Decoder

a1a2a3…01 110 11…

Page 32: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

32

Timely CompressionA Status Updating Problem

Encoder

01 110 11…

FIFObuffer

Rate Rbit pipe Decoder

a1 a2 a3 a4 a5 a6 a7 …

ta1 a2 a3a4 …

Page 33: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

33

Timely CompressionHuffman Block Coding Example

Channel Rate R

Sta

tus

Ag

e

High rate pipe:

use small blocks

Low rate pipe:

Use large blocks

Page 34: WINLAB 1 Roy Yates ECE/WINLAB, Rutgers NSF Workshop on Ultra Low Latency Wireless Networks March 26, 2015 Age of Information Status Updating Systems and

WINLAB

34

Summary

• Information Age Minimization– Match the load to the network/system

• “Rate” is an input for controlling delay – Redesign the system

• Give priority to timely updates• Packet Management

• Ultra Low Latency Networks– Sub 1ms latency applications?

• Better Theory for Network Latency