Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks

Richard Draves, Jitu Padhye, Brian Zill

Microsoft Research

Self-Organizing Neighborhood Networks

Internet

Gas Station

Bus Stop

Mesh Router 2

End Device(Guest to Router 1)

Mesh Router 1

Mesh End Device

EXIT

Mesh Zone

Mesh Router 3

(Internet TAP)

Mesh Router 5

Mesh Router 7

90

101

206

• Key Properties– No network engineer– Very little mobility– Energy not a concern

• One challenge:network capacity

• Our approach: multiple radios

Results

• Ad-hoc routing at layer 2.5 works well

• Link quality is important, but not all metrics are created equal

• Multiple radios provide significant capacity improvement if the routing utilizes channel-diversity, data rate, loss rate

(Please see our SIGCOMM & Mobicom papers for more details.)

Layer 2 vs Layer 3

• Layer 2 (link layer): like ethernet switches− Limited to single link technology+ Supports multiple protocols (IPv4, IPv6, IPX)+ Preserves link abstraction

• Layer 3 (network layer)+ Supports multiple link technologies− Limited to single network protocol− Link-local mechanisms don’t work

• DHCP, RA/RS

Our Approach: Routing at Layer 2.5

• A virtual link-layer+ Supports multiple link technologies

+ Supports IPv4, IPv6 etc unmodified

+ Preserves the link abstraction

+ Agnostic to choice of ad-hoc routing algorithm

Ethernet 802.11 802.16

Mesh Connectivity Layer (with LQSR)

IPv4 IPv6 IPX

Mesh Connectivity Layer (MCL)

• Virtual ethernet adapter– Virtual ethernet addresses– Multiplexes heterogeneous

physical links– Physical links need not be

ethernet

Ethernet

MCL

Payload:TCP/IP,

ARP,IPv6…

Packet Format

Link-Quality Source Routing (LQSR)

• Source-routed link-state routing protocol– Derived from DSR– Part of Mesh Connectivity Layer (layer 2.5)– Supports link-quality modules

• Both on-demand/proactive mechanisms– Route Discovery– Route Maintenance– Metric Maintenance

LQSR Metric Support

• HOP: shortest-path routing– closest to DSR

• RTT: round-trip time latency

• PktPair: packet-pair latency

• ETX: expected transmission count

• WCETT: designed for multiple radios

Multi-Radio Routing

• Previous metrics (HOP, ETX) not suitable for multiple radios per node– Do not leverage channel, range, data rate diversity

• Weighted Cumulative Expected Transmission Time– Weight links according Expected Transmission Time (ETT)

• Takes link bandwidth and loss rate into account

– Combine link ETTs into Weighted Cumulative ETTs (WCETT) • Takes channel diversity into account

– Incorporated into source routing

WCETT: Combining link ETTs

All hops on a path on the same channel interfere– Add ETTs of hops that are

on the same channel

– Path throughput is dominated by the maximum of these sums

Need to avoid unnecessarily long paths - bad for TCP performance - bad for global resources

Given a n hop path, where each hop can be on any one of k channels, and tuning parameter β:

j iij

jkj

n

ii

ETTX

where

XETTWCETT

channel on is hop

11

max)1(

Select the path with min WCETT

Testbed

• 23 nodes in building 113• Cheap desktop

machines– HP d530 SF

• Two radios in each node– NetGear WAG or WAB– Proxim OriNOCO– Cards can operate

in a, b or g mode.

205

201

204

203

210

226

220

227

221

225

224

206

211

207

208

209

219

215

216

218

217

214

223

Appro

x. 6

1 m

Approx. 32 m

TCP Throughput Test

• Select 100 sender-receiver pairs at random (out of 23x22 = 506)– 2-minute TCP transfer

• Two scenarios:– Baseline (Single radio):

• NetGear cards in 802.11a mode• Proxim OFF

– Two radios• NetGear cards in 802.11a mode• Proxim cards in 802.11g mode

• Repeat for shortest-path, ETX, WCETT

ResultsMedian Throughput of 100 transfers

16011379

1155

1508

844

2990

0

500

1000

1500

2000

2500

3000

3500

WCETT ETX Shortest Path

Th

rou

gh

pu

t (K

bp

s)

Single Radio

Two Radios

WCETT uses 2nd radio better than ETX or shortest path.

Two-Radio Throughput CDF

0

0.2

0.4

0.6

0.8

1

0 2 4 6 8 10 12 14 16

Throughput (Mbps)

Fra

ctio

n o

f C

on

nec

tio

ns

wit

h L

ow

er

Th

rou

gh

pu

tHOPETXWCETT

Better

Two-Radio Path Length vs Throughput

WCETT

0

2

4

6

8

10

12

14

16

18

0 1 2 3 4 5 6 7

Average Path Length (Hops)

Th

rou

gh

pu

t (M

bp

s)

ETX

0

2

4

6

8

10

12

14

16

18

0 1 2 3 4 5 6 7

Average Path Length (Hops)

Th

rou

gh

pu

t (M

bp

s)

WCETT Improvement by Path Length

0

20

40

60

80

100

120

140

1 2 3 4 >=5

Path Length (with 2 radios)

Per

cen

tag

e Im

pro

vem

ent

in M

edia

n

Th

rou

gh

pu

t

Conclusions

• Ad-hoc routing at layer 2.5 works well

• Link quality is important for performance

• Previous routing metrics do not work well in heterogeneous multi-radio scenarios

• WCETT improves performance by making judicious use of additional capacity and channel diversity provided by the 2nd radio