1

Multi-Channel Wireless Networks:Capacity, Protocols, and Experimentation

Nitin VaidyaUniversity of Illinois at Urbana-Champaign

In collaboration withPradeep Kyasanur

Chandrakanth ChereddiVartika Bhandari

Keynote at WINLAB Research Review (Rutgers University), November 15, 2006

2

Multi-hop Wireless Networks

g Wireless paradigms:Single hop versus Multi-hop

g Multi-hop networks:Mesh networks, ad hoc networks, sensor networks

3

Wireless Capacity

g Wireless capacity limited

g In dense environments, performance suffers

g How to improve performance ?

4

Improving Wireless Capacity

g Exploit physical resources,and resource diversity/multiplicity

g Exploiting diversity requires appropriate protocols

g Examples …iBeamforming antennasiExploiting infrastructure (hybrid networks)iBetter spatial reuse via rate/power/carrier sense adaptationiMulti-channel

5

This Talk

Utilizing multiple channels in wireless networks

g Capacity bounds

g Protocol design

g Experimentation (Net-X testbed)

6

Multiple Channels

g Typically, available frequency spectrum is split into multiple channels

26 MHz 100 MHz 200 MHz 150 MHz

2.45 GHz915 MHz 5.25 GHz 5.8 GHz

3 channels 8 channels 4 channels

250 MHz 500 MHz 1000 MHz

61.25 GHz24.125 GHz 122.5 GHz

Large number of channels available

7

Channel Model

g c channels available

g Bandwidth per channel W

Channel 1

Channel 2

Channel c

8

Radio Interfaces

g Switching between channels may incur delay

g An interface can only use one channel at a time

Channel 1

Channel c

9

Interface Model

g Reducing hardware cost allows formultiple interfaces

g m interfaces per node: Typical values of msmall

1

m

10

Channel-Interface Scenarios

g Scenario 1: m = c One interface per channel

11

Common case

11

m = c c = m

With sufficient hardware

11

Channel-Interface Scenarios

g Scenario 2: m < c A host can only be onsubset of channels

1

c

1

m m

Likelyscenario

m+1c–m unused channelsat each node

12

Multi-Channel Mesh

g How to best utilize multiple channelsin a mesh networkwith limited hardware ?

13

Need for New Protocols

g When m < ciHow to assign channels to interfaces at each host?iWhen to switch an interface among channels?iHow to select good routes?

1,2

Some channels not used

A B C

D 1,2

1,21,2

Network poorly connected

A B C

D

1,3

2,4

1,2 3,4

c = 4, m = 2

14

Outline

Utilizing multiple channels in wireless networks

g Capacity bounds

g Protocol design

g Experimentation (Net-X testbed)

15

Capacity Analysis

g How does capacity improve with more channels ?

g How many interfaces needed to best use c channels ?

iClearly, m = c suffices for maximum performance

iNot always possible to have c interfaces

16

Worst Case

g Worst case capacity is m/c fraction of thebest-case

A B Channel data rate = W

c interfaces: cW throughput

m interfaces: mW throughput

g What about other scenarios ?

17

Capacity = ?[Gupta-Kumar]

g Random source-destination pairs among randomly positioned n hosts in unit area, with n ∞

18

Capacity = ?

g λ = minimum flow throughputg Capacity = n λ

19

Capacity Constraints

g Capacity constrained by availablespectrum bandwidth

g Other factors further constrainwireless network capacity …

20

Connectivity Constraint[Gupta-Kumar]

g Need routes between source-destination pairsPlaces a lower bound on transmit power

Not connected Connected

A D AD

21

Interference Constraint[Gupta-Kumar]

g Interference among simultaneous transmissionsg Limits spatial reuse

AB

> r

DC

r

22

Capacity of Wireless Networks[Gupta-Kumar]

g When c = m,

Capacity increases linearly with channels

11

c = mm = c

W

W

network capacity α

23

Capacity

g What if fewer interfaces ?

1

m

1

c

mm+1

24

Interface Constraint

g Throughput is limited by number of interfaces in a neighborhood

N nodes in the “neighborhood”

total throughput ≤ N * m * W

Interface, a constrained resourcein addition to spectrum, time and space

25

Destination Bottleneck Constraint

g A node may be destination of multiple flows

g Node throughput shared by all the incident flows

Df incomingflows

Node throughput T ≤ m*W

Per-flow throughput = T / f

26

Mutlti-Channel Network Capacity[Kyasanur-Vaidya]

Ratio c/mBest caseCapacity α c W

27

Mutlti-Channel Network Capacity

Ratio c/m

Connectivity and interference

Interference andinterface bottleneck

Interface anddestination bottlenecks

28

Possible to benefit fromlarger spectrum despite

channel switching constraints

Constrained Channel Switching[Bhandari-Vaidya]

g “Untunable radios”[Petrovic05] restricted to use a subset of channels(vary across devices)

or, spatially correlated channel assignments

g A device can communicate directly with only a subset of in-range nodes that share usable channel

(1, 2)

(2, 3)

(1, 3)

(2, 5)(7, 8)

(6, 7)(3, 6)

(5, 6)

(4, 5)

29

Outline

Utilizing multiple channels in wireless networks

g Capacity bounds

g Protocol design

g Experimentation (Net-X testbed)

30

Towards Protocols

Analysis Practical constraints

Insights on protocol design Software architecture

Protocol design & implementation

31

Insights from Analysis (1)

g Static channel allocation does not yield optimal performance in general

g Must dynamically switch channels

g Need protocol mechanisms for channel switching

A

C

BChannel 1

2 D3

32

Insights from Analysis (2)

g Optimal transmission range function of

density of nodes and

number of channels

Intuition:# of interfering nodes ≈ # of channels

33

Insights from Analysis (3)

g Routes must be distributed within a neighborhoodg This is not necessary in single channel networks

AB

C

D

E

F

A

B

C

D

E F

Multi-Channel (m<c)Optimal strategy

Single Channel (m=c=1)Optimal strategy

34

Insights from Analysis (4)

g Channel switching delay potentially detrimental, but may be hidden with

icareful scheduling – create idle time oninterfaces between channel switches

iadditional interfaces

g Idle interfaces can switch channelswithout penalizing performance

35

Insights from Analysis (5)

Channel Assignment

g Need to balance load on channels

g Local coordination in channel assignment helpful

36

Practical Constraints

g Legacy MAC : 802.11g Non-trivial topology discovery costsg Non-trivial channel switching cost

Design decisions:

g Multi-channel awareness above MAC layerg Channel management on two timescales

37

Channel Management: Timescale Separation[Kyasanur-Vaidya]

g Routing: Longer timescales

iMulti-channel aware routeselection metrics

g Interface management:Shorter timescales

iDynamic channel assignmentiInterface switching

Link

Network

Transport

PhysicalLayer

Upper layers

802.11

38

Channel Assignment

g Interfaces may be switched or kept fixed

g Classification:i Static strategy: All interfaces of a node fixedi Dynamic strategy: All interfaces of a node can switchi Hybrid strategy: Some interfaces fixed, others switch

g We use a hybrid strategy requiring at least two interfaces per node

39

A B C

Channel Assignment

g Two interfaces much better than oneg Hybrid channel assignment: Static + Dynamic

Fixed (ch 1)

Switchable

Fixed (ch 2)

Switchable

Fixed (ch 3)

Switchable12 3 2

Channel assignment locally balanced

40

Routing Approach

g Legacy routing protocols can be operated over our interface management layer

iDoes yield significant benefits from multiple channeliDoes not consider cost of channel switching

g An alternative

iDevelop a channel-aware metric(aware of channel diversity and switching costs)

41

44

4

Selecting Channel Diverse Routes

g Most routing protocols use shortest-hop metriciNot sufficient in most multi-channel networks

A needs route to C

Route A-B-C better

Prefer channel diverse routes

3A B C

D E F

2

1 3 4

4 2

42

1 43

Impact of Switching Cost

g Interface switching cost has to be considerediA node may be on multiple routes, requiring switching

Select routes that do not require frequent switching

23

2

Route A-B-C in use

D needs route to F

Route D-E-F better4

A B C

D E F

2 4 2

43

CBR – Random topology(50 nodes, 50 flows, 500m x 500m area)

0

2

4

6

8

10

12

14

16

1 2 3 4 5 6 7 8 9 10

Nor

mal

ized

thro

ughp

ut

Topology number

(2,2)(2,5)(5,5)

(2,12)(12,12)

(m,c)

44

Outline

Utilizing multiple channels in wireless networks

g Capacity bounds

g Protocol design

g Experimentation (Net-X testbed)

45

Net-X Testbed[Kyasanur-Chereddi-Vaidya]

g Linux 2.4

g Two 802.11a radiosper mesh node (m = 2)

g Legacy clients with1 radio

g c = 5 channels

Soekris 4521 Net-X source code tobe released soon

46

Net-X Testbed

Two radio mesh node

Internet gateway node Single radio unmodifed client

Single radio mesh node

47

New Kernel Support

g O.S. support needed to choose channels based on destination A

B

CCh. 1

Ch. 2

Next hop not equivalent to a wireless interface id

g Phy-aware routing not supported traditionally

g In general, need a “constraint” specificationfor desired channel(s), antenna beamform,power/rate, … to be used for the next hop

48

New Kernel Support

A

B

C

Ch. 1

Ch. 2D

Ch. 3

g Multi-channel (phy-aware)broadcast support needed

g Channel switching from user space has high latency: frequent switching from user space undesirable

49

New Kernel Support

g Interface management needs to be hidden from“data path”

– Buffering packets for different channels– Scheduling interface switching

Packet to B

Packet to C

Ch. 2

Ch. 1

Packet to C arrives

buffer packet

Interface switchesto channel 1

50

Net-X Architecture

Multi-Channel Routing,Channel Assignment

Interface and ChannelAbstraction Layer

IP Stack

InterfaceDevice Driver

User Applications

ARP

InterfaceDevice Driver

g Abstraction layer simplifies use of multiple interfaces

Implemented by extending Linux “bonding driver”

51

Channel Abstraction Module

g Unicast Component: iAllows choosing channels based on destination

g Broadcast Component: iMulti-channel broadcast support

g Queueing and Scheduling Component:iQueue packets if interface is not immediately availableiSchedule interface switching

52

Ongoing Work

g Testbed deployment ongoing(approximately 30 nodes when fully deployed)

g Extensive measurements planned

Partialtestbed

view

CSLSouthwing

53

Research Opportunities

g > 2 interfaces • Channel assignment issues• Multi-path routing• Protocol simplification

g QoS differentiation

g Interpreting results in time-domain:Channel = power-save duty cycle

g Cognitive radios: Dynamically determine channel availability

g Constrained channel switching

54

Conclusions

g Insights from the analysis useful in protocol design

g Significant performance benefits usingmany channels despite limited hardware

g Implementation requires new O.S. support•Net-X source to be made public soon

g Significant research opportunities remain

55

Thanks!

www.crhc.uiuc.edu/wireless