Wireless Mesh Networks Victor Bahl http://research.microsoft.com/~bahl

1
Wireless Mesh NetworksVictor
Bahlhttp//research.microsoft.com/bahl

• Lecture 1
• MSR India Summer School on Networking

June 15, 2007
2
These lecture notes are for educating. Feel free
to incorporate these slides in your presentations
but please cite the source on each borrowed slide
and as a courtesy to the author please inform him
of such use. Do not post a copy of these slides,
or slides derived from these on a web site
without the authors written permission.
Notice

• The contents of this deck may change without
  notice

.
3
Foreword

• Mobile ad hoc networking and mesh networking is a
  thriving area of research. The number of
  solutions results are simply too large to cover
  in a short lecture.
• This is not a lecture on (1) wireless
  communications (2) MAC protocols, (3) PHY Layer
  techniques and (3) multi-hop routing protocols.
• This is quick talk about what I know about
  building mesh networks. Exhaustive deep
  treatment of all existing results is not
  provided.
• These notes are an attempt to describe the main
  problems the general idea behind some of the
  promising solutions. At the end of this lecture
  you should have a reasonably good understanding
  of the state-of-art in mesh networking.

.
4
Topics not covered in this lecture

• Due to lack of time I was not cover several
  important research results that you should also
  be aware of.
• Some of these are
• Modulation and PHY techniques like OFDM, Analog
  Network Coding etc.
• Adaptive antenna technologies like MIMO, beam
  forming, etc.
• TCP enhancements (for mesh networking)
• Routing protocols (there are hundreds)
• Multicast routing and group communications
• Topology control and power management
• Standards including IEEE 802.11s,
• Security and Management
• Directional MACs etc

.
5
Roadmap

• Mesh Networking Applications
• Basics of Radio Frequency Communications
  (already covered by Dr. Ramjee)
• Multi-hop Wireless Networking
• Historical background
• Challenges Mesh networking with 802.11
• Handling the Challenges
• Capacity Enhancement Calculation
• MMAC, SSCH, BFS-CA, HMCP, MUP, Network Coding,
  conflict graphs, .
• Routing Protocols Link Quality Metrics
• RFC 2501, RFC 3626, RFC 3684, RFC 3561, RFC
  4728, ETX, LQSR, EXoR, HWMP, ETT, WCETT
• Security Network Management
• Mesh Deployments Discoveries Innovations
• MSRs Mesh, MITs RoofNet, IITs DGP, Rices TFA,
  UMASSs DieselNet, UCSBs Mesh,
• Madcitys Mesh, JHUs SMesh
• Mesh Networking Standards
• IEEE 802.11s (IETF standards covered previously)
• References

Will not cover, tutorial notes available on
request
.
6
Mesh Networking Applications
7
Wireless Mesh Networking

• Definition
• A wireless mesh network is a peer-to-peer
  multi-hop wireless network in which participant
  nodes connect with redundant interconnections and
  cooperate with one another to route packets.
• Unlike Mobile Ad hoc NETworks (MANETs) where
  routings node are mobile, in mesh networks
  routing nodes are stationary.
• Mesh nodes may form the network's backbone. Other
  non-routing mobile nodes ("clients") may connect
  to the mesh nodes and use the backbone to
  communicate with one another over large distances
  and with nodes on the Internet

.
8
Characteristics of a Mesh Network
Classic Hub Spoke Network
Mesh Network

• Can grow organically
• Does not require infrastructure support
• Is fault tolerant
• Requires distributed management
• Offers higher capacity (via spatial diversity
  power management), but
• Too many nodes ? shared bandwidth may suffer due
  to interference
• Too few nodes ? route maintenance is difficult
  disconnections possible
• Identity and security management is a challenge

.
9
The Mesh Networking World
Internet
Broadband
Neighbourhood
Home Mesh
Mesh Node
Mesh Node
Mesh Node
Mesh Node
Traditional Last Mile Territory
.
10
Scenario 1 Broadband Internet Access
Internet
Backbone
Middle Mile
Last Mile

• Cost of middle and last miles make physical wired
  infrastructure not an option in rural areas and
  many countries
• Equipment capital cost
• The scale of touching / maintaining so many
  endpoints
• The physics of running cable large distances over
  unfriendly terrain
• Political, social and territorial implications
• Wireless mitigates these issues but introduces
  others
• Range
• Bandwidth
• Spectrum availability
• Cost maintenance issues of new hardware /
  standards
• Mesh networking makes wireless workable
• Range bandwidth addressed by shorter links
• Cost maintenance addressed by building on
  commodity standards

.
11
Scenario 2 A Community Mesh Network
Organic Participants own the equipment and the
network
.
12
Community Mesh Network Applications

• Shared broadband Internet access
• Neighborhood watch (e.g. video surveillance)
• Shared media content (e.g. neighborhood DVR)
• Medical emergency response
• Neighborhood eBay (garage sales, swaps)
• Billboards (babysitter/service recommendations,
  lost cat, newsletter)
• Bits produced locally, gets used locally
• Social interaction
• Distributed backup
• Internet use increased social contact, public
  participation and size of social network.
  (social capital - access to people, information
  and resources)
• Prof. Keith N. Hampton (author of Netville
  Neighborhood Study)
• URL http//www.asanet.org/media/neville.html

.
13
Scenario 3 Home Mesh

• Extend Access Point (AP) coverage
• Better spectrum (re)use ? greater capacity
• Automatic discovery, plug-and-play networked home
  devices
• AV equipment (Cameras, TV, DVD, DVR,
  satellite/cable)
• Phones (Cellular and POTS)
• Traditionally disassociated smart devices (PDAs,
  AutoPC)
• Home infrastructure items (Light switches, HVAC
  controls)

.
14
Scenario 4 Blanket City-wide Wireless Coverage

• Philadelphia picks Earthlink for City Wireless,
  TechNew World, October 5, 2005
• San Francisco Keeps Pushing City Wide WiFi, CNET
  News.com, August 17, 2005
• San Francisco Mayor Gavin Newsom wants to make
  Wi-Fi coverage in the city as ubiquitous as the
  fog that blankets its neighborhoods.
• Wi-Fi Hits the Hinterlands, BusinessWeek Online,
  July 5, 2004
• Who needs DSL or cable? New mesh technology
  is turning entire small towns into broadband hot
  spots, Rio Rancho N.M., population 60,000, 500
  routers covering 103 miles2
• NYC wireless network will be unprecedented,
  Computerworld, June 18, 2004
• New York City plans to build a public safety
  wireless network of unprecedented scale and
  scope, with a capacity to provide tens of
  thousands of mobile users
• Rural Areas need Internet too! Newsweek, June 7,
  2004 Issue
• EZ Wireless built the country's largest
  regional wireless broadband network, a
  600-square-mile Wi-Fi blanket, and activated it
  this February, Hermiston, Oregon, population
  13,200, 35 routers with 75 antennas covering 600
  miles2
• Mesh Casts Its Net, Unstrung, January 23, 2004
• Providing 57 miles2 of wireless coverage for
  public safety personnel in Garland Texas
• PCCW takes Wireless Broadband to London, The
  Register, September 2, 2005
• Prices for the service in UK start from 10 /
  month for 256 Kbps to 18 /month for 1 Mbps

.
15
Scenario 5 All-Wireless Office

• Older buildings
• For small offices (100 PCs)
• Rapid deployment
• Low cost
• Short-term offices
• Not a replacement for wire

• No wires
• No switches
• No APs

.
16
Scenario 6 Spontaneous Mesh

• Definition
• A temporary ad-hoc multihop wireless network for
  exchanging voice, video or data, for
  collaboration in a locally distributed
  environment, when no permanent infrastructure or
  central control is present. Usually between
  portable wireless devices.
• 1. Peer Calling Party Lines
• P2P calling within local groups conferences,
  events, school campus,

2. Public Safety Fire and rescue teams need
ad-hoc communication at incident sites
3. Real Time Advisory Drivers need traffic
information and advisories generated in real time
.
17
Grass Roots Mesh Deployments

• Academia
• The Roofnet Project (MIT, USA) -
  http//pdos.csail.mit.edu/roofnet/doku.php
• 802.11 mesh network for broadband IA in cities
• The CITRIS TIER Project (UC Berkeley, USA) -
  http//tier.cs.berkeley.edu/
• Technology and Infrastructure for emerging
  regions
• The Digital Gangetic Plains Project (IIT Kanpur,
  India) - http//www.iitk.ac.in/mladgp
• 802.11-based low-cost Networking for rural
  India
• The TFA Project (Rice University, USA) -
  http//taps.rice.edu/index.html
• Technology for All Project
• .
• Community Mesh Networks
• Community Network Movement - http//www.scn.org/co
  mmnet/
• Seattle Wireless - http//www.seattlewireless.net/
  
• Champaign-Urbana Community Wireless Network -
  http//www.cuwireless.net/
• Kingsbridge Link, U.K. - http//www.kingsbridgelin
  k.co.uk/
• .

.
18
Industry Breakdown
Infrastructure Based
Infrastructure-less
SkyPilot, QualNet (Flarion), Motorola (Canopy)
IRoamAD, Vivato, Arraycomm, Malibu Networks,
BeamReach Networks, NextNet Wireless, Navini
Networks, etc.
Meshnetworks Inc.,Radiant Networks, Invisible
Networks, FHP, Green Packet Inc., LocustWorld,
etc.
Architecture effects design decisions on Capacity
management, fairness, addressing routing,
mobility management, energy management, service
levels, integration with the Internet, etc.
.
19
Industry Deployment Scenarios
http//www.unstrung.com/insider/
March 2005, Source Unstrung Insider
.
20
What about WiMAX?

• IEEE 802.16d for developing/rural use (.16e
  targets mobile scenarios)
• Still needs market momentum around hardware
  optimisation size, power, efficiency and most
  importantcost
• WiMAX as a last-mile solution?
• In low-density areas, WiMAX requires high-power
  towers or lots of towers (gt cost goes up)
• In NLOS environments, range impacts bandwidth
  through reduced modulation
• WiMAX CPE expensive in next 3-5 years (
  150-250)
• WiMAX feeding a mesh can be a good solution
• Mesh extends WiMAX tower reach
• Mesh simplifies the financials by greatly
  reducing equipment cost
• Mesh is robust and deal with network vagaries

.
21
WiMAX Mesh

• WiFi Meshes can add value to WiMAX in several
  ways
• Reduce CPE costs
• Extend range of WiMAX tower without compromising
  speed
• Replace high-price WiMAX towers with cheaper mesh
  nodes

16QAM
16QAM
16QAM
16QAM
16QAM
8PSK
WiMAX Only
WiMAX with Mesh
QPSK
FSK
A
A
.
22
Multi-Hop Wireless Networking
23
Historical Perspective

• Packet Radio Network (PRNET), 1972-1982
• Band 1718.4-1840 MHz Power 5 W Range 10 km
  Speed100-400 Kbps, Addressing Flat Routing
  Distance Vector Scale 50
• Survivable Adaptive Networks (SURAN), 1983-1992
• Band 1718.4-1840 MHz Power 5 W Range 10 Km
  Speed 100-400 kbps, Addressing Hierarchical
  Routing Distance Vector Scale 1000 (Low cost
  packet radio)
• Global Mobile Information Systems (GLOMO),
  1995-2000
• e.g. NTDR, Band 225-450 MHz Power 20 W Range
  11-20 Km Speed 300 kbps, Addressing Flat
  Routing Link-state / 2-level clusters Scale
  400
• IETF Mobile Ad Hoc Networks (MANET) Working
  Group, 1997
• RFC 2501 (Eval), RFC 3561 (AODV), RFC 3626
  (OLSR), RFC 3684 (TBRPF), Drafts DSR, DYMO,
  Multicast, OLSRv2
• MSR Mesh Networking Project (2002 2005)
• IEEE 802.11s Working Group, 2004 -

PRNET Van
.
24
Challenge Mesh Networking with IEEE 802.11
25
The MAC Problem Packets in Flight Example
RTS
RTS
RTS
RTS
RTS
2
3
4
5
7
8
9
1
11
10
6
CTS
CTS
Backoff window doubles ?
2 packets in flight! Only 4 out of 11 nodes are
active.
.
26
Throughput Internet Gateway Example
Internet
RTS
RTS
RTS
CTS
Backoff window doubles
Backoff window doubles
.
27
The Scheduling Problem

• yes

• yes

If future traffic is not known, which one do you
schedule first?
.
28
The Fairness Problem
1
2
Jinyang-MobiCom-2001

• Information Asymmetry
• A C do not have the same information
• C knows about flow 1 (knows how to contend)
• A does not know about flow 2
• Flow 2 always succeeds, Flow 1 suffers
• When RTS/CTS is used
• As packets are not acknowledged by B
• A times out doubles its contention window
• When RTS/CTS is not used
• As packet collide at B, but Flow 2 is succesful
• A times out double its contention window
• Downstream links suffer

Gambiroza-MoiCom-2004
.
29
The Fairness Problem (2)
ITAP
Camp-DC-2005

• Location closest to gateway gets the more packets
• Nodes farthest from the gateway get very little
  bandwidth and can get starved
• Possible solution Rate control on each node with
  fairness in mind
• Need topology traffic information to calculate
  fair amount
• Global vs. distributed solution

.
30
The Fairness Problem (3)

• MAC attempts to provide fairness at packet level
  not flow level
• Capture phenomena
• Winner of competing flows has a higher chance of
  winning contention again
• Different levels of interference at different
  links (different neighborhood)
• Highly interfered flows can be drowned

Nandagopal-MobiCom-2000
Qiu-MSRTR-2003
Flow1 Flow2 Flow3 Flow4 Flow5
2.5 Mbps 0.23 Mbps 2.09 Mbps 0.17 Mbps 2.55 Mbps
Active area of research - MACAW, WFQ, DFS,
Balanced MAC, EBF-MAC, PFCR, .
.
31
The Path Length Problem

• Experimental Setup
• 23 node testbed
• One IEEE 802.11a radio per node (NetGear card)
• Randomly selected 100 sender-receiver pairs (out
  of 23x22 506)
• 3-minute TCP transfer, only one connection at a
  time

Impact of path length on throughput
If a connection takes multiple paths over
lifetime, lengths are byte-averaged Total 506
points.
.
32
The Collision Problem
Robert Morriss Rooftnet MSR Mesh Summit 2004
Presentation
Multi-hop collisions cut b/w by about 2x
Actual Roofnet b/w is often much lower
Expected multi-hop b/w based on single-hop b/w
33
The Node Density Problem
Round trip delay versus node density
A new 100Kbps CBR connection starts every 10
seconds, between a new pair of nodes. All nodes
hear each other.
.
34
The Power Control Problem
Tight power control reduces interference and
increases throughput

• A B do not detect RTS/CTS exchange between C
  D
• B does not detect data transmission from D to C
• Bs transmission to A results in packet collision
  at C

.
35
The Power Control Problem (2)

• Tight power control reduces interference
  increases overall throughput
• But it also disconnects the network. So whats
  the right power control algorithm?

.
36
The Capacity of Mesh Nodes
What is the maximum achievable capacity of a mesh
network with N nodes?
Gupta-IEEEIT-2000

• Optimal Case
• Nodes are optimally located, destinations are
  optimally located
• Traffic patterns are fixed
• Optimally spatio-temporal scheduling, routes,
  ranges for each transmission
• As each node obtains bits/sec

• Average Case
• Randomly located nodes and destinations
• Traffic pattern are random
• Each node chooses same range
• Each node obtains bits/sec

.
37
The Capacity Calculation Problem

• Gupta and Kumar 2000
• Theorem for stationary ad hoc nodes in the worst
  case traffic scenario
• Determines asymptotic, pessimistic bounds on
  performance
• Every node in the mesh is active (either
  transmitting or receiving)
• Does not answer
• What is the capacity of a mesh which is using
  multiple channels, directional antennas, tight
  power control?

.
38
What is the Real Capacity of a Chain?
but the radios interferance range is gt radios
communication range
Source
Destination
1
2
3
4
5
6
With Ideal MAC, Chain Utilization 1/3
With interferences, Chain Utilization 1/4
Jinyang-MobiCom-2001
.but this is achievable only with optimum
scheduling and optimum offered load!, with
random scheduling and random load, utilization
1/7 !
.
39
Routing Problem Which to Choose?Unicast Ad Hoc
Multi-hop Routing Protocols

• ABR (Associativity-Based Routing Protocol)
• AODV (Ad Hoc On Demand Distance Vector)
• ARA (Ant-based Routing Algorithm)
• BSR (Backup Source Routing)
• CBRP (Cluster Based Routing Protocol)
• CEDAR (Core Extraction Distributed Ad hoc
  Routing)
• CHAMP (CacHing And MultiPath routing Protocol)
• CSGR (Cluster Gateway Switch Routing)
• DART (Dynamic Address Routing)
• DBF (Distributed Bellman-Ford)
• DDR (Distributed Dynamic Routing)
• DNVR (Dynamic Nix-Vector Routing)
• DSDV (Dynamic Destination-Seq. Dist. Vector)
• DSR (Dynamic Source Routing)
• DSRFLOW (Flow State in the DSR)
• DYMO (Dynamic Manet On-Demand)
• FORP (Flow Oriented Routing Protocol)
• FSR (Fisheye State Routing)
• GB (Gafni-Bertsekas)

• LANMAR (LANdMARk Routing Protocol)
• LAR (Location-Aided Routing)
• LBR (Link life Based Routing)
• LCA (Linked Cluster Architecture)
• LMR (Lightweight Mobile Routing)
• LQSR (Link Quality Source Routing)
• LUNAR (Lightweight Underlay Network Ad hoc
  Routing)
• MMRP (Mobile Mesh Routing Protocol)
• MOR (Multipoint On-demand Routing)
• MPRDV (Multi Point Relay Distance Vector)
• OLSR (Optimized Link State Routing)
• OORP (OrderOne Routing Protocol)
• DREAM (Distance Routing Effect Algorithm for
  Mobility)
• PLBR (Preferred Link Based Routing)
• RDMAR (Relative-Distance Micro-discover Ad hoc
  Routing)
• Scar (DSR and ETX based)
• SSR (Signal Stability Routing)
• STAR (Source Tree Adaptive Routing)
• TBRPF (Topology dissemination Based on
  Reverse-Path Forwarding)

.
40
The Path Selection Problem

• Several link quality metrics to select from
• Hop count
• Round trip time
• Packet pair
• Expected data transmission count incl.
  retransmission
• Weighted cumulative expected transmission time
• Signal strength stability
• Energy related
• Link error rate
• Air Time
• Which to select? We still dont have a
  interference-aware metric! We still dont know
  how to measure interference..

.
41
Baseline comparison of Metrics Single Radio Mesh

• Experimental Setup
• 23 node testbed
• One IEEE 802.11a radio per node (NetGear card)
• Randomly selected 100 sender-receiver pairs (out
  of 23x22 506)
• 3-minute TCP transfer, only one connection at a
  time

Median path length HOP 2, ETX 3.01, RTT
3.43, PktPair 3.46
ETX performs the best
Draves-MobiCom-2004
.
42
Baseline Comparison of Metrics Two Radio Mesh
Draves-SIGCOMM-2004

• Experimental Setup
• 23 node testbed
• Randomly selected 100 sender-receiver pairs (out
  of 23x22 506)
• 3-minute TCP transfer
• Two scenarios
• Baseline (Single radio)
• 802.11a NetGear cards
• Two radios
• 802.11a NetGear cards
• 802.11g Proxim cards

Median path length HOP 2, ETX 2.4, WCETT 3
WCETT utilizes 2nd radio better than ETX or
shortest path
.
43
But with different traffic pattern.

• Trace Capture
• 1 workstations connected via Ethernet
• Traces captured during 1-month period
• Trace Replayed
• Testbed of 22 mesh computers in office
  environment
• 2 IEEE 802.11a/b/g cards per computer

Erickson-MobiSys-2006
.
44
The Multicast Problem

• A multicast group is defined with a unique group
  identifier.
• Nodes may leave or join the group anytime
• In wired networks physical network topology is
  static
• In ad hoc multi-hop wireless networks physical
  topology can change often
• Need to Integrate with unicast routing protocols
• Many proposals Tree-based, Mesh-based,
  Location-based which one to use?

• - ABAM (On-Demand Associatively-Based
  Multicast) - FGMP (Forwarding Group Multicast
  Protocol)
• - ADMIR (Adaptive Demand-Driven Multicast
  Routing) - LAM (Lightweight Adaptive Multicast)
• - AMRIS (Ad hoc Multicast Routing utilizing
  Increased id-numberS) - MAODV (Multicast AODV)
• - DCMP (Dynamic Core Based Multicast Routing) -
  MCEDAR (Multicast CEDAR)
• - AMRoute (Adhoc Multicast Routing) - MZR
  (Multicast Zone Routing)
• - CAMO (Core-Assisted Mesh Protocol) - ODMRP
  (On-Deman Multicast Routing Protocol)
• - CBM (Content Based Multicast) - SPBM
  (Scalable Position-Based Multicast)
• - DDM (Differential Destination Multicast) -
  SRMP (Source Routing-based Multicast Protocol)
• - DSR-MB (Simple Protocol for Multicast and
  Broadcast using DSR) -
• -

.
45
The Interference Detection Problem

• When two systems operate on overlapping
  frequencies, there exists a potential for harmful
  interference between them
• Performance degradation on both systems
• Conflict graph is determined by the Interference
  Graph
• To determine the Interference Graph, require
• Knowledge of packet transmission from nodes that
  are not visible
• Knowledge of physical location of nodes within
  the network
• Knowledge of whether or not multiple
  transmissions increase ot decrease interference?
• Interference Graph can change
• as rapidly as the environment
• when a node leaves or join the network

.
46
The Transport Layer Problem

• Majority of the Internet traffic is TCP
• Packet losses delays in wireless can occur due
  to
• Environmental fluctuations resulting link
  failures
• Stochastic link performance due to rapidly
  changing error rates
• CSMA/CA assumes loss is due to congestion and
  back-offs
• TCP assumes packet losses are due to congestion
• Times out when no ACK is received
• Invokes slow start, when instead the best
  response would be to retransmit lost packets
  quickly
• RTT calculation can change as rapidly as the
  environment (link) changes
• Can we solve this problem without changing the
  end-to-end semantics?

.
47
The Security Problem

• Two type of attackers
• External malicious node (no crypto keys)
• Compromised node (attacker captures legitimate
  node and reads out all cryptographic information)
• Attacks
• Selfish behavior, do not forward other nodes
  packets
• Denial of Service (DoS)
• Jamming
• Resource consumption attack
• Routing disruption (e.g. Wormhole attack)
• Inject malicious routing information
• Ongoing Research
• Possible solutions SEAD, Ariadne, SRP,
  CONFIDANT,

Hu-MobiCom-2002
Bucheggar-MobiHoc-2002
.
48
The Spectrum Etiquette ProblemLocal behavior
affects Global Performance!
Doesnt care
Packets get dropped!
.
49
Consequently we..

• Must Increase Range and Capacity
• Single radio meshes built on 802.11 technologies
  are not good enough. We must extend the range of
  radios we must understand the achievable
  capacity in an ideal wireless mesh and we must
  build technology to approach this capacity?
• Must Improve Routing Performance
• Routing protocols based on shortest-hop are
  sub-optimal. We must build a routing protocol
  that adapts quickly to topology changes,
  incorporates wireless interference and link
  quality.
• Must Provide Security and Fairness
• Is it possible to ensure fairness and privacy for
  end-users and security for the network? We must
  ensure that no mesh nodes starves and that the
  mesh guards itself against malicious users.
• Must Provide Self Management
• An organic network should be both
  self-organizing and self managing? To what extent
  can we remove the human out of the loop?
• .
• Must Develop a Resilient Framework for
  Applications
• In a environmentally hostile environment, we must
  provide a framework for applications to work
  robustly.

.
50
Handling the Challenges
.
51
Strategies for increasing Capacity

• Strategy 1 Use all available channels
• Avoid spectrum waste
• Strategy 2 Improve modulation, reception, and
  coding
• Today 2.5 bits/Hz (.11g), Soon 4.5 bits / Hz
  (.11n)
• Network coding
• Strategy 3 Improve spatial reuse by reducing
  interference
• Fine grain transmit power control
• (Steerable) directional antennas and directional
  MACs
• Strategy 4 Navigate around harmful interference
• Interference aware least cost routing

Will not cover
.
52
Strategy 1 Multi-Channel Communications

• Goal
• Assign n non-interfering channels to n pair of
  nodes such that n packet transmissions can occur
  simultaneously.

Knobs
Single Channel Multiple Channels
Single Radio Today ?
Multiple Radio X ?
.
53
Single Radio Multiple Channels (SR-MC)

• Distributed Use a modified RTS/CTS sequence to
  negotiate channels
• Problem
• How does the sender know which channel the
  receiver is listening on?
• Solutions
• Receive on all channels simultaneously
• Simplest solution but too costly - will not
  consider here
• Use a dedicated rendezvous channel
• Use a synchronized hopping protocol
• Provide multiple rendezvous opportunities
• Centralized Compute channel assignments using
  global knowledge
• Scope of Coverage
• We will cover schemes that work on commodity
  radios only

.
54
Packets-in-Flight Example Revisited

• Negotiating Channel with RTS / CTS

RTS (C1,C3,C7)
RTS (C3,C5,C7,C11)
C2
C2
C1
C11
C1
2
3
4
5
7
8
9
1
11
10
6
CTS (C11)
CTS (C1, C7)
10 nodes are active, 5 packets in flight, 150
improvement!
.
55
Note Hidden Terminal Multi-Channel Case

• Let C1 be the rendezvous channel
• ? can hear traffic on C1 only, doesnt hear the
  CTS from ß consequently doesnt know anything
  about traffic on C6 (d is too far to hear
  anything from ß)

So-MobiHoc-2004
Possible solution Use multiple radios
.
56
Implementation Option for SR-MC
Buffer packets, switch between channels
Chandra-INFOCOM-2004
Channel switching speed Today - 5
milliseconds Possible - 80 microseconds
Application Layer
User-level
Kernel-level
TCP/IP, Network Stack
802.11 Device Driver Switching logic
Firmware
Packets for C6
Packets for C11
Packets for C1
802.11 hardware
57
Multi-Channel Medium Access Control (MMAC)
Idea Periodically rendezvous on a fixed channel
to decide the next channel

• Divide time into beacon intervals
• Divide a beacon interval into two phase
• Negotiation Phase All nodes switch to a
  pre-defined common channel and negotiate the
  channel to use
• Transfer Phase Once a channel is selected, the
  source receiver switch to this channel and data
  transfer occurs during this phase

So-MobiHoc-2004

• Issues
• Requires tight clock synchronization
• Packets to multiple destinations can incur high
  delays
• Congestion on the common channel
• Common channel goes bad, everything goes bad
• Not able to handle broadcasts

.
58
Slotted Seeded Channel Hopping (SSCH)

• Divide time into slots
• At each slot hop to a different channel
• Nodes hop across channels to distribute traffic
• Senders and receivers probabilistically meet
  exchange schedules
• Senders loosely synchronize hopping schedule to
  receivers
• Characteristics
• Distributed every node makes independent choices
• Optimistic exploits common case that nodes know
  each others channel hopping schedules
• Traffic-driven nodes repeatedly overlap when
  they have packets to exchange

Bahl-MobiCom-2004
.
59
SSCH Rendezvous
Divide time into slots switch channels at
beginning of a slot
New Channel (Old Channel seed) mod (Number of
Channels) seed is from 1 to (Number of Channels -
1)

(1 2) mod 3 0
Seed 2
3 channels E.g. for 802.11b Ch 1 maps to 0 Ch 6
maps to 1 Ch 11 maps to 2
A
0
2
1
0
2
0
1
1
B
Seed 1
0
1
2
0
1
2
0
1
(0 1) mod 3 1

• Enables bandwidth utilization across all
  channels
• Does not need control channel rendezvous

.
60
SSCH Syncing Seeds

• Each node broadcasts (channel, seed) once every
  slot
• If B has to send packets to A, it adjusts its
  (channel, seed)

Seed
2
2
2
2
2
2
2
2
2
A
0
2
1
0
2
0
1
1
2
3 channels
B wants to start a flow with A
B
2
0
2
1
0
2
1
1
0
2
Seed
1
2
2
2
2
2
2
1
Stale (channel, seed) info simply results in
delayed syncing
.
61
Using all Available Channels with SSCH
In current IEEE 802.11 meshes
Only one of 3 pairs is active _at_ any given time
.
62
SSCH Performance
100 nodes, IEEE 802.11a, 13 channels, every flow
is multihop
Avg. per node Throughput
Total System Throughput
SSCH
SSCH
IEEE 802.11a
IEEE 802.11a
Significant capacity improvement when traffic
load is on multiple separate flows
.
63
How many Channels can we really use?
Banerjee-SIGMETRICS-2006

• IEEE 802.11b,g partitions the allocated 83.5
  MHz spectrum into 11 channels
• Only channels 1, 6 and 11 are mutually
  non-overlapping
• Butusing only the orthogonal channels may waste
  spectrum

.
64
How many Channels can we really use?

• Can we use more channels by using
  partially-overlapped channels?
• Caution may increase interference and cause more
  harm than good
• Need an appropriate model to capture
  interference-effects and make correct choices

65
Overlapped Channels do work!
Link A, Channel 1
ChSep 5
ChSep 2
ChSep 1
Distance (X-axis)
ChSep 0
Link B, Channel Y

• Minimum distance between links for different
  choices of channel separation
• For every channel separation there is a minimal
  distance
• Model-based algorithmic approach for channel
  assignment in wireless mesh networks

.
66
Notes on Single Radio Multiple Channels

• Single radio solutions can be applied to
  multi-radios nodes since in most cases the number
  of channels is greater than the number of radios
  in the node.
• Compared to multi-radio solutions, single radio
  solutions are power efficient but power is not
  the primary concern in most mesh networks
• Single radio solutions are less costly than
  multi-radio solutions but radios are fairly
  inexpensive
• Switching speeds and mute-deaf-time is a problem
  in single radio solutions but switching speeds
  are being reduced dramatically
• When distance between nodes is large, need not
  restrict operation to non-overlapping channels
  only

so now lets look at multi-radio solutions
.
67
Single Node Multiple Radio - Interference Study

• Question
• Do two radios operating on non-overlapping
  channel interfere?
• Experimental setup

HOP 1
TCP
A
B
6 separation between B C radios
TCP
C
D
HOP 2
.
68
802.11a/g Interference Results

• Same channel or channel separation of 4 causes
  46 - 49 reduction in overall throughput

802.11a link causes a 22 reduction in overall
throughput, and a 63 reduction in throughput
on the 802.11g link. Surprise 802.11g does
not affect 802.11a

• Implications
• Interference even when radios are placed 6
  apart is significant
• Significant RF hardware shielding work is needed

.
69
Single Node Multiple Radios

• Lets assume we can build mesh-boxes with
  enough separation / shielding between radios that
  performance does not suffer.
• Then interesting problem to consider
• (1) How should we assign channels to each
  interface?
• Dont want to cause network partitions
• (2) Which interface should we send the packet
  on?
• State-of-art metrics (hop count, ETX, SRTT,
  packet-pair) are not suitable for multiple radio
  / node. As they do not leverage channel, range,
  data rate diversity.

.
70
Multiple Radios - Multiple Channels

• Options to consider
• Static Assignment
• One channel / radio for all time
• Suboptimal use of spectrum
• Some routes may be suboptimal
• Dynamic Assignment (all SR-MC strategies apply)
• Channels assigned to match traffic patterns
  and/or to reduce interference
• Interference patterns can change,
• network may get disconnected
• Hybrid Assignment
• One channel to one radio for all time, for all
  other radios, channels are assigned dynamically
  to match traffic patterns and/or reduce
  interference

.
71
Static Assignment (1)
2 radios / node
Draves-MobiCom-2004
All nodes use common set of channels
11
1
B
C
A
11
11
Suboptimal use of spectrum
1
1
11
1
1
D
E
F
11
11
.
72
Static Assignment (2)
2 radios / node, 4 channels
Raniwala-Infocom-2005
Different nodes use different channels
56
52
B
C
A
Some routes may be suboptimal (e.g. B-gtF)
52
52
60
60
D
E
F
64
60
.
73
Dynamic Assignment
N radios / node M channels N lt M Interfaces
can switch channels as needed
B
C
A
Coordination may be needed before each
transmission
D
E
F

• MMAC
• SSCH
• BFS-CA

See section on single radio multiple channels
.
74
Breadth First Search Channel Assignment (BFS-CA)
Dynamic Assignment
Ramachandran-Infocom-2006

• Goals
• External interference can severely degrade mesh
  performance
• Measure avoid external interference
• Internal interference between mesh links should
  be avoided
• Assign orthogonal channels to any two interfering
  mesh links
• Nodes periodically estimate surrounding
  interference levels
• A radio per-node monitors 802.11 data and control
  traffic
• Channels ranked from least interfered to most
  interfered
• Ranking sent to centralized channel assignment
  server
• Distributed channel sensing and channel
  assignment
• can break network connectivity
• A radio per-router is tuned to a common channel
  to
• ensure connected mesh

.
75
BFS-CA Interference-Aware Channel Assignment

• Multi-radio conflict graph models interference
  between mesh links
• Breadth first search algorithm selects channels
  for mesh radios
• Significant performance improvement over static
  channel assignment in the presence of varying
  interference levels
• Issues
• Interference patterns can change rapidly
• Dependence on existance of traffic patterns to
  determine interferance
• Incorrect channel assignment possible

.
76
Hybrid Multichannel Control Protocol (HMCP)
Hybrid Assignment

• Each node has two interfaces (1 fixed, 1
  switch-able)
• Connectivity is maintained all channels used
• Every node picks a channel as its fixed channel
• Different nodes use different fixed channels
• Sender tunes its switchable interface to
  receivers fixed channel to send packets
• Once a connection is made, there may not be a
  reason to switch channels again for that
  particular flow.

Kyasanur-WCM-2006
.
77
HMCP Channel Selection

• Challenge Nodes in a neighborhood should use
  different fixed channels
• Fixed Channel Selection
• On startup pick a random fixed channel
• Periodically send a hello pkt. containing fixed
  channel 1-hop neighbors info. on all channels
  (using the switchable interface)
• Maintain a NeighborTable containing fixed
  channels being used by neighbors
• Select the channel with fewest nodes as a
  candidate
• Use 2-hop neighbor information
• Change fixed channel to candidate channel
  probabilistically to avoid oscillations
• Issues
• High overhead for broadcast packets

.
78
Which Interface should we send the packet on?
Adya-BroadNets-2004

• A Simple Approach For every transmission select
  the interface with the best channel transmit
  on it
• Multi-Radio Unification Protocol (MUP)
• Pros
• - Locally optimizes use of available spectrum
• - Does not require changes to routing protocols
  or application-level software
• - Interoperates with legacy hardware
• - Does not require global topology information
• Cons
• For one-hop ad hoc works great. For meshes need
  metrics that combine link selection metrics into
  a path selection metric (will see)

.
79
MultiRadio Unification Protocol (MUP)
Illustration of Channel Switching

• Goal
• Allow nodes with multiple radios to locally
  optimize use of available spectrum and hence
  increase capacity
• Operation
• Set the network interface cards on different
• frequency channels
• Periodically monitor channel / Link
• quality on each interface
• Select the interface with the best channel
• and transmit packets

Ch. 0
0
1
Ch. 1
Does not require global topology information
Adya-BroadNets-2004
.
80
MUP in a Neighborhood
Mesh formation among 35 randomly selected houses
252 houses in a Seattle neighborhood (Green Lake
Area)
Web surfer
Routes via RFC 3561 (AODV)
40-50 reduction in delay compared to a
one-radio network
ITAP
.
81
Why MUP is not enough?

• MUP is a link metric not a path metric. Routing
  protocols that use MUP do not
• Leverage channel diversity
• A two hop path with hops on different channels is
  better than a path with both hops are on the same
  channel.
• Leverage range and data rate diversity
• A path with two 6 Mbps hops is better than a path
  with a single 1 Mbps hop. MUP will take the 1Mbps
  path.
• ..but a path with four 6 Mbps hops is worse than
  a path with a single 2 Mbps hop. MUP and metrics
  like ETX may take the four-hop path, depending on
  delay loss rate.
• Note Striping protocols are not enough
• Packet-level striping results is packet
  reordering, and hence poor TCP throughput
• Flow-level striping requires a routing algorithm!
• MUP may work, but only if radios have identical
  range.
• Bottom Line Need a routing protocol / metric
  that takes bandwidth, loss rate, and channel
  diversity into account.

.
82
about routing in mesh networks
83
The Routing Problem

• Why not simply use traditional routing protocols
  (RIP/OSPF/etc)?
• Network topologies are dynamic due to router
  mobility environmental fluctuations
• Dynamic topology may prevent routing protocol
  convergence
• Many links are redundant (routing updates can be
  large)
• Periodic updates may waste bandwidth batteries
• Computed routes may not work due to
  unidirectional links
• Wireless makes routing protocols easy to attack
• Link quality, spectrum utilization and
  interferences are uniquely important for path
  selection

.
84
Desirable Qualitative Properties

• Distributed operation
• Loop-freedom
• Demand-based operation
• Proactive operation
• Attack resistant Secure
• Sleep period operation (friendly to power
  management)
• Unidirectional link support / asymmetric link
  support
• Implementation
• Layer 3 - traditional network layer / IP layer
• Interoperable internetworking capability and
  consistency over a heterogeneous networking
  infrastructure.
• Layer 2.5
• Agnostic of IPv4 or IPv6 issues and can
  incorporate link quality measures more easily
• Capable of handling multiple wireless wired
  networking technologies

Corson-RFC2501-1999
.
85
Routing and Addressing

• Many choices, which is the best one?
• Flat addressing - Each node runs the routing
  protocol nodes address is independent of its
  location (e.g. PRNET, TORA, DSR, AODV,..)
• Clustering Only cluster heads run routing
  protocol addressing is flat and independent of
  nodes location (NTDR, CEDAR,..)
• Hierarchical Only cluster heads run routing
  protocol, a nodes form subnets, each node
  acquires address of its subnet (SURAN).
• Many protocols to consider
• See next
• Multiple path routing
• Many choices MSDR, AOMDV, AODV-BR, APR, SMR,
  ROAM, .

.
86
Bucketizing Routing Protocols

• Proactive (periodic)
• Each node maintains route to each other network
  node (Global state)
• Routes are determined independent of traffic
• All topology changes propagated to all nodes
• Periodic routing advertisements (neighbor
  discovery is beacon based)
• Generally longer route convergence time
• Examples Distance vector and link state (DSDV,
  OLSR, TBRPF)
• Reactive (on-demand)
• Actions driven by data packet requiring delivery
• Source builds route only when needed by
  flooding (Route Discovery)
• Maintain only active routes (Route Maintenance)
• Pro Typically less overhead, better scaling
  properties
• Cons Route acquisition latency
• Examples DSR, AODV

87
Conventional Wisdom

• Proactive protocols perform best in networks with
  low to moderate mobility, few nodes and many data
  sessions
• E.g. OLSR (RFC 3626), TBRPF (RFC 3684)
• Reactive protocols perform best in
  resource-limited, dynamic networks where nodes
  are mobile. Tradeoff routing overhead for
  start-up delay
• E.g. AODV (RFC 3561), DSR (IETF Draft)

88
Popular Taxonomy
.
89
Mapping Protocols to Taxonomy
.
90
Common Metrics for Comparing Routing Protocols

• Route Acquisition Time
• Time required to establish route(s)
• Routing Overhead
• Total number of routing packets transmitted (for
  discovery maintenance) for a fixed amount of
  transfer over multiple hops with random node
  mobility.
• Path Optimality / End-to-End Throughput
• TCP UDP data throughput and delay
• Previously path optimality was measured in terms
  of the difference between the number of hops a
  packet took to reach its destination and the
  length of the shortest path that physically
  existed through the network when the packet
  originated. However, the hop count metric has
  been overshadowed by other link metrics when it
  came to end-to-end throughput
• Packet Delivery Ratio
• Ratio between the number of packets originated by
  the application layer and the number of packets
  received by the sink at the final destination

Broch-MobiCom-1998
.
91
Context for Comparing Metrics

• Network size
• Measured in terms of the number of nodes
• Network connectivity
• Measured in terms of avg. node degree (i.e. the
  avg. number of neighbors)
• Topological rate of change
• Speed with which a network's topology changes
• Link capacity
• Effective link speed in bps, after accounting
  for losses due to multiple access, coding,
  framing, etc.
• Fraction of unidirectional links
• Transmission ranges of radios may be different
• Traffic patterns
• Long-lived versus bursty non-uniform
• Mobility
• Described in terms of dwell time, movement
  direction, speed etc.
• Fraction and frequency of sleeping nodes

Corson-RFC2501-1999
.
92
Link / Path Selection Metrics

• Min. hop count results in lower-quality links

Incorporate metric into routing protocols
93
Path Selection Metrics

• Link Metric Assign a weight to each link
• Prefer high bandwidth, low-loss links
• RTT, Packet Pair, ETX
• Metrics such as shortest path, RTT, Packet Pair,
  ETX etc. do not leverage channel, range, data
  rate diversity
• Path Metric Combine metrics of links on path
• Prefer short, channel-diverse paths
• WCETT

.
94
Expected Transmission Count (ETX)Link Selection
Metric for Single Radio Meshes
Couto-MobiCom-2003

• Advantages
• Explicitly takes loss rate into account
• Implicitly takes interference between successive
  hops into account
• Low overhead
• Disadvantages
• PHY-layer loss rate of broadcast probe packets is
  not the same as PHY-layer loss rate of data
  packets
• Broadcast probe packets are smaller
• Broadcast packets are sent at lower data rate
• Does not take data rate or link load into account

• Each node periodically broadcasts a probe
• The probe carries information about probes
  received from neighbors
• Each node can calculate loss rate on forward (Pf)
  and reverse (Pr) link to each neighbor
• Selects the path with least total ETX

.
95
Expected Transmission Time (ETT) Link Selection
Metric for Single Radio Meshes

• Given
• Loss rate p
• Bandwidth B
• Mean packet size S
• Min backoff window CWmin
• Takes bandwidth and loss rate of the link into
  account

96
Weighted Cumulated ETT (Combine link ETTs)Link
Selection Metric for Multi-Radio Meshes

• Given a n hop path, where each hop can be on any
  one of k channels, and two tuning parameters, a
  and b

Path throughput is dominated by the max of the
sum of ETTs of path links on the same channel
Sum of ETTs of all links on the path -
Favors short paths
Draves-MobiCom-2004
Sum of ETTs of all links on the path that are on
the same channel
Select the path with min WCETT
Takes bandwidth, loss rate and channel diversity
into account
97
Path Length and ThroughputWhich metric is best?
(Wireless Office Study)
Eriksson-MobiSys-2006

• Experimental Setup
• 23 node testbed
• Randomly selected 100 sender-receiver pairs (out
  of 23x22 506)
• 3-minute TCP transfer (transmit as many bytes as
  possible in 2 minutes, followed by 1 minute of
  silence)

For 1 or 2 hop the choice of metric doesnt
matter
.
98
Comparison of MetricsWireless Office Scenario
Eriksson-MobiSys-2006
23 node indoor testbed. Two radios (both 802.11a)
per node. 11 active clients, 4 servers.
Heavy Office Traffic 1 hour, 308 sessions, 587.5
MB total
Light Office Traffic 1 hour, 415 sessions, 19.72
MB total
Relatively light traffic means performance is
okay for all metrics. WCETT does better under
heavy load (worst case delay)
.
99
Summarizing

• Many routing protocols to choose from
• Protocols that take link quality into account
  show most promise
• A link quality metric that incorporates
  interference is still needed
• Adaptive protocols that change behavior in
  different environments might be best

.
100
Additional Areas of Research
101
Active Areas of Research

• Analytical tools for calculating mesh capacity
• Flow-level and packet-level fairness
• Network management automatic diagnosis of
  faults
• Network coding for capacity improvement
• Routing with directional antennas / routing for
  network coding
• Supporting VoIP video traffic over meshes
• Inexpensive software steerable directional
  antennas
• Smart medium access control
• Meshing with cognitive radios
• Multi-spectral meshes
• Delay tolerant meshing
• Usage scenarios

.
102
Thanks!For prior work updates, check
outhttp//research.microsoft.com/nrg/
Complete tutorial notes available on my web
site http//research.microsoft.com/bahl
Q/A
103
References

• Papers are categorized under subject area.
  Duplication is possible because some papers
  include more than one problem/solution pair. This
  is not a exhaustive list. List is not exhausted
  (was prepared at least 2 years ago)

104
References - Testbeds

• UMASSs DieselNet
• Zhao-MASS-2006 Wenrui Zhao, Yang Chen, Mostafa
  Ammar, Mark Corner, Brian N. Levine, and Ellen
  Zegura. Capacity Enhancement using Throwboxes in
  DTNs, IEEE Intl Conf on Mobile Ad hoc and Sensor
  Systems (MASS), Oct 2006.
• Partan-WUWNet-2006 Jim Partan, Jim Kurose,
  Brian N. Levine, A Survey of Practical Issues in
  Underwater Networks. ACM Intl Wkshp on Underwater
  Networks (WUWNet), September 2006
• Jun-ACHANTS-2006 Hyewon Jun, Mostafa Ammar,
  Mark Corner, Ellen Zegura. Hierarchical Power
  Management in Disruption Tolerant Networks with
  Traffic-Aware Optimization, ACM CHANTS.
  September, 2006.
• Burgess-INFOCOM-2006 John Burgess, Brian
  Gallagher, David Jensen, and Brian N. Levine.
  MaxProp Routing for Vehicle-Based
  Disruption-Tolerant Networks, IEEE INFOCOM, April
  2006.
• Burns-ICRA-2006 Brendan Burns, Oliver Brock,
  and Brian N. Levine. Autonomous Enhancement of
  Disruption Tolerant Networks, IEEE Intl Conf on
  Robotics and Automation (ICRA), May 2006.
• Burns-INFCOMM-2005 Brendan Burns, Oliver Brock,
  and B.N. Levine. MV routing and capacity building
  in disruption tolerant networks., IEEE INFOCOM,
  March 2005.
• Hanna-ICNP-2003 Kat Hanna, Brian N. Levine, and
  R. Manmatha. Mobile Distributed Information
  Retrieval For Highly Partitioned Networks, IEEE
  ICNP, November 2003.

.
105
References - Testbeds

• IITKs Digital Gangetic Plains
• Chebrolu-MobiCom-2006 Kameswari Chebrolu,
  Bhaskaran Raman, and Sayandeep Sen, Long-Distance
  802.11b Links Performance Measurements and
  Experience, 12th Annual International Conference
  on Mobile Computing and Networking (MOBICOM), Sep
  2006, Los Angeles, USA.
• Raman-MobiCom-2005 Bhaskaran Raman and
  Kameswari Chebrolu, Design and Evaluation of a
  new MAC Protocol for Long-Distance 802.11 Mesh
  Networks, 11th Annual International Conference on
  Mobile Computing and Networking (MOBICOM),
  Aug/Sep 2005, Cologne, Germany.
• Bhagwat-HotNets-2003 Pravin Bhagwat, Bhaskaran
  Raman, and Dheeraj Sanghi, Turning 802.11
  Inside-Out, Second Workshop on Hot Topics in
  Networks (HotNets-II), 20-21 Nov 2003, Cambridge,
  MA, USA.
• MITs RoofNet
• Briket-MobiCom-2005 John Bicket, Daniel Aguayo,
  Sanjit Biswas, and Robert Morris, Architecture
  and Evaluation of an Unplanned 802.11b Mesh
  Network, ACM MobiCom 2005.
• Biswas-SIGCOMM-2005 Sanjit Biswas and Robert
  Morris, Opportunistic Routing in Multi-Hop
  Wireless Networks, ACM SIGCOMM 2005
• Aguayo-SIGCOMM-2004 Daniel Aguayo, John Bicket,
  Sanjit Biswas, Glenn Judd, Robert Morris,
  Link-level Measurements from an 802.11b Mesh
  Network, SIGCOMM 2004, Aug 2004
• Couto-MobiCom-2003 Douglas S. J. De Couto,
  Daniel Aguayo, John Bicket, Robert Morris, A
  High-Throughput Path Metric for Multi-Hop
  Wireless Routing, ACM Mobicom 2003

.
106
References - Testbeds

• MSRs Mesh Network
• Qiu-CCR-2006 Lili Qiu, Paramvir Bahl, Ananth
  Rao, Lidong Zou, Troubleshooting Wireless
  Meshes, ACM Computer Communications Review 2006
• Eriksson-MobiSys-2006 Jacob Eriksson, Sharad
  Agarwal, Paramvir. Bahl, Jitendra Padhye,
  Feasibility Study of Mesh Networks for
  All-Wireless Offices, ACM/USEN