Throughput and Coverage Improvement in Wireless Mesh Networks

1
(No Transcript)
2
Existing Wireless Networks

• High cost of infrastructure-based networks
• All access points (AP) require wired connection
• Expensive in developed as well as rural areas
• Makes deployment difficult
• Lack of service guarantees in MANETs
• Multihop wireless without infrastructure
• Also seeding problem

3
An Alternative Wireless Mesh Networks

• TAP nodes are wirelessly connected to each other
• TAP nodes are centrally operated
• Only gateway TAPs are directly connected to
  Internet
• Alternative to DSL and cable

4
Thesis Contributions

• Throughput improvement for best-effort traffic
• Traffic-aware routing metrics
• Coverage improvement
• Deployment of low-cost booster TAPs (bTAPs)
• Scheduling scheme for coexistence
• The two solutions can complement each other

5
A Motivating Example

• In the absence of flow f1
• Route S-C-D-E gives higher throughput
• Hop count metric selects S-A-B
• With flow f1 present
• Route S-A-B gives higher throughput
• Existing flow f1 consumes capacity of D-E
• Need to account for interference from existing
  traffic

6
Related Work

• QoS-related approaches
• Calculate spare bandwidth
• Perform admission control on new flow
• Reserve bandwidth
• Best-effort approaches
• Minimum hop count metric
• Expected Transmission Count (ETX) metric
• Expected Transmission Time (ETT) metric
• Weighted Cumulative ETT (WCETT) metric
• Interference-Aware Resource Usage (IRU) metric

7
Problems with Existing Metrics

• They are traffic-unaware
• They are immune to network load
• Packet loss probability only traffic dependent
  component
• Especially problematic in multi-user scenarios

8
Requirements for a Routing Metric

• A metric should account for
• Packet loss probability of each link
• Modulation rate of each link
• Interference caused by existing flows
• The MAC Layer Share (MLS) metric
• The AVAIL model-based metric

9
MAC Layer Share (MLS)

• Network activity at some node

10
The MLS Metric

• Dependent on
• MLS of source node
• Modulation rate of link
• Link loss probability of link

11
MLS Metric Example
12
The AVAIL Metric

• Analytical estimation of per link throughput
• Maximum additional input rate for each link
• Based on a model of IEEE 802.11
• Michele, Theodoros, and Knightly
• Extended the model to handle multiple receivers
• Joint work with Theodoros, Michele and Knightly
• Dynamic parameters in model taken from real
  measurements

13
From Link Throughput to Route Throughput

• Minimum of the maximum throughput for each clique
• Source TAP throttles flow at estimated throughput

14
The Routing Protocol

• Combination of link state routing and source
  routing
• Periodic link state updates sent by each TAP
• Each node has a view of the topology
• Each source selects route based on the sources
  own view
• Purely on-demand routing would be unsuitable
• Finds the best route at time of route discovery
• Does not find better route later, if one appears
• Using source routing guarantees no loops

15
Evaluation

• Compare traffic-aware vs. traffic-unaware metrics
  
• ns-2 simulation
• Models IEEE 802.11 CSMA / CA
• RTS / CTS was not used
• Single channel at 11 Mbps rate
• 150 m transmission range
• Interference up to 212 m
• UDP traffic
• Studied new flow added to existing traffic
• Studied with omni-directional antennas and with
  directional antennas

16
The Manhattan-Grid Topology

• Synthetic topology
• 196 nodes
• 10 gateways

17
The Chaska Topology

• Modeled after the Chaska, MN network deployed by
  Tropos
• 194 nodes, 14 gateways
• I could only use the abstract topology
• Real transmission range not known

18
Existing Traffic

• 100 random sources each send to nearest gateway
• Load on each gateway is 30 100 of a maximum
  gateway load
• Equally distributed among all flows ending at
  that gateway
• Example
• If maximum gateway load is 2 Mbps
• And a particular gateway is loaded with 50 of 2
  Mbps
• And 5 flows end at that gateway
• Then each flow sends at (0.5 ? 2 Mbps) / 5 0.2
  Mbps

19
Single Additional Flow

• Allow the 100 flows to run for 30 seconds
• Start one new flow to gateway
• Randomly choose a node not already a source
• Each simulation run thus has 101 flows
• Source selects gateway according to different
  metrics
• Source chooses best possible gateway
• MLS, AVAIL, MLS-T, ETX, IRU
• Repeat for 50 simulation runs
• Compare throughput achieved by the additional flow

20
The MLS-T Metric

• The MLS metric overestimates throughput
• The MLS-T selects same route as MLS
• After route selection, model is used for
  throughput estimation

21
Overall Results Manhattan-Grid Topology

Maximum Gateway Load 2 Mbps
Maximum Gateway Load 3 Mbps
22
Overall Results Chaska Topology

Maximum Gateway Load 1.5 Mbps
Maximum Gateway Load 2 Mbps
23
Thesis Contributions

• Throughput improvement for best-effort traffic
• Traffic-aware routing metrics
• Coverage improvement
• Deployment of low-cost booster TAPs (bTAPs)
• Scheduling scheme for coexistence
• The two solutions can complement each other

24
Related Work

• MADF ad hoc overlay added to fixed
  infrastructure
• Overlay operates on separate channel
• Clients in overloaded cell use this overlay
• ICAR balances traffic load between cells
• Stationary relays borrow channels from
  non-congested cells
• UCAN similar approach to serve clients with weak
  signal
• Ad Hoc City city-wide ad hoc network
• Network comprised of moving fleet of buses
• Cellular DSR routing protocol

25
Coverage Improvement Motivation

• Signal strength degrades rapidly with distance
• Node B receives much weaker signal than does node
  A

26
Coverage Improvement New Idea

• Deploy low-cost booster TAPs (bTAPs)
• At most one bTAP on route from TAP to client node
• Transmission coordination difficult over multiple
  hops
• Hard to justify cost and complications of extra
  hops

27
Design Goals for Low-Cost bTAPs

• Should have a lower antenna tower than TAP
• Should have a single radio
• Communicate both with TAP and client nodes
• Can use multiple antennas
• Should use omni-directional antenna with clients
• Should be comparable in cost to client nodes

28
System Model Idealized Cell

• Idealized cell containing 6 sectors
• Aggressive frequency reuse pattern
• Similar to reuse pattern used in cellular
  networks
• Narrower sectors require costlier equipment
• Each sector uses separate orthogonal channel
• TAP has 6 radios and 6 directional antennas

29
Placement of bTAPs

• Two bTAPs required to cover edge of sector
• The bTAPs operate on orthogonal channels
• One bTAP per sector

30
Simultaneous Scheduling

• Nodes A and B can simultaneously communicate on
  same channel

31
Scheduling A Possible Map

• Divide into three types of periods
• Backhaul
• Dedicated
• Simultaneous
• Red sector is independent of green sector

32
Effective Data Rate

• Time to transmit B bytes of data
• Effective rate

33
Evaluation

• Experiments using MATLAB simulations
• Compute coverage improvement
• Sector area that can communicate at ? QPSK ½ (6
  Mbps)
• Compute average sector throughput improvement

34
Simulation Parameters
35
(1,6,6) Frequency Reuse Patterns
36
Multi-Cell Scenario with bTAPs
37
Simultaneous Scheduling Regions Downlink
38
Dedicated Scheduling Regions Downlink
39
Improvement in Coverage (1,6,6) Reuse
40
Improvement in Sector Throughput (1,6,6)
41
Thesis Contributions

• Throughput improvement for best-effort traffic
• Traffic-aware routing metrics
• Coverage improvement
• Deployment of low-cost booster TAPs (bTAPs)
• Scheduling scheme for coexistence
• The two solutions can complement each other

42
The Complete System
43
A Possible End-to-End Schedule
44
Simultaneous Schedule Sector and Inter-TAP

• All but one period can be simultaneous

45
Conclusion Future Work

• Extension to multiple orthogonal channels
• Simple extension
• The effect of TDMA-based MAC protocol
• The effect of different routing metrics on TCP
• A hybrid routing protocol
• Proactive and reactive
• Testing the metrics in real deployments
• Example The TFA deployment in Houston

46
Conclusion Thesis Summary

• Novel traffic-aware routing metrics
• MLS, AVAIL, and MLS-T metrics
• Outperform traffic-unaware metrics
• Chaska topology severely limited by interference
• Directional antennas improve performance
• Deploy bTAPs to improve coverage of TAP nodes
• Improved coverage from 75 to greater than 95
• Improved sector throughput by up to 33
• Improvements higher in (1,6,6) than in (1,3,6)
• Complete system architecture
• Scheduling map to allow coexistence

47
DATA / ACK nowledgements

• Thesis committee members
• Dave, Ed, and Eugene
• Technical collaborators
• Theodoros, Michelle
• J, Shankar
• Grants and gifts
• NSF, NASA, Rice, Schlumberger
• Rice community members
• Department
• OISS

48
DATA / ACK nowledgements

• Group members
• Santa, Shu, Khoa, Yanjun
• Jorjeta, Yih-Chun
• Friends and well wishers
• Dhruv, Rajnish, Vinod, Priyank, Pradeep,
• Baddy, Squash, Cricket, FoYM, ISAR members
• Dr. Winston, Liz, Dr. Jenkins, Dr. Ware
• Family
• Parents, brother, relatives
• Deepanwita (PhT)

49
(No Transcript)
50
Tracking MLS

• Components are easy to track at MAC layer
• Current device drivers do not export API
• No reason why APIs cannot be exported

51
MLS Limitation

• Contention window is not part of MLS definition
• Highly dynamic
• Limited benefit from incorporation into metric
• Contention window not used in TDMA-based
  MAC(e.g., IEEE 802.16)

52
End-to-End Throughput
53
From Link Throughput to Route Throughput
10 pkts/s
54
Throughput Improvement Motivation

• Route S-C-D-E gives higher throughput
• Hop count metric select S-A-B

55
Throughput, Preset Maximum 1.5 Mbps
56
Delay, Preset Maximum 1.5 Mbps
57
Route Length, Preset Maximum 1.5 Mbps
58
Directional Antenna Pattern

• 3 dB beamwidth of 300

59
With Directional Antenna Chaska

• Preset Maximum 1.5 Mbps
  2 Mbps

60
Throughput, Preset Maximum 3 Mbps
61
Delay, Preset Maximum 3 Mbps
62
Route Length, Preset Maximum 3 Mbps
63
Throughput, Preset Maximum 2 Mbps, CO
64
Delay, Preset Maximum 2 Mbps, CO
65
Route Length, Preset Maximum 2 Mbps, CO
66
Throughput, Preset Maximum 3 Mbps, MD
67
Delay, Preset Maximum 3 Mbps, MD
68
Route Length, Preset Maximum3 Mbps, MD
69
Multiple Additional Flows

• Allow the 100 flows to run for 30 seconds
• Start 50 additional flows, 1 every 5 seconds
• Each of these flows lives for 50 seconds
• Throttle new flow to half of estimated throughput
• Search route at most once every second
• Shift to new route if at least 10 higher
  throughput

70
Overall Results Multiple Flows, Manhattan

Maximum Gateway Load 2 Mbps
Maximum Gateway Load 3 Mbps
71
Overall Results 4Mbps, Manhattan

Single additional flow
Multiple additional flows
72
Summary

• Chaska topology severely limited by interference
• Directional antennas marginally increase average
  throughput
• Improvement of up to 12 by using traffic-aware
  metrics
• Traffic-aware metrics outperform traffic-unaware
• MLS, AVAIL, MLS-T perform better than ETX, IRU
• Up to 50 higher average throughput
• Directional antennas provide high improvement
• 28 - 45 improvement for traffic-unaware metrics
• 25 - 50 improvement for traffic-aware metrics

73
Erceg-Greenstein Model
74
Terrains in Erceg-Greenstein Model

• Terrain A Hilly/Moderate-to-heavy tree density
• Terrain B Hilly/Light tree density or
  Flat/Moderate-to-heavy tree density
• Terrain C Flat/Light tree density

75
Directional Antenna Model
76
Alternate Placement of bTAPs

• Range of bTAP radio has to be large
• Requires high bTAP antenna height
• Higher bTAP transmission power
• Causes high inter-sector co-channel interference
• Multiplexing required at bTAP

77
Analyzing Coverage Improvement

• Single-sector model is insufficient
• Cannot capture best-case scenario
• Nodes of best pair might lie in differentsectors
• Twin-sector model allows moreflexibility in
  choosing pairs

78
Simulation Parameters Transmission Rates

• Representative values from IEEE 802.16 (WiMAX)

79
Frequency Reuse Patterns
(1,6,6) Reuse Pattern
(1,3,6) Reuse Pattern
80
Simulation Parameters
81
Directional Antenna Pattern

• 3 dB beamwidth of 30

82
Multi-Cell Scenario without bTAPs
83
Without bTAPs no Lognormal Fading
84
Without bTAPs with Lognormal Fading
85
Variation with Pathloss and Fading
86
Variation with Pathloss and Fading
87
Results with (1,3,6) Reuse Pattern

• Original
  With bTAPs

88
Effective Rates Downlink
89
Results with (1,3,6) Reuse Pattern
90
Summary

• Substantial coverage improvement with bTAPs
• Coverage improves from 75 to 95 for (1,6,6)
• Coverage improves from
• Sector throughput increases by up to 33 for
  (1,6,6)
• Gains sufficient to offset costs of radio
  resources required for forwarding user traffic
  via bTAPs

91
Wireless Statistics

• WiFi hotspots 129,000 over 130 countries
• United States 41,810, UK 16,360, Germany 12,934
  France 10,703, South Korea 9,415 Japan 6,316
  Taiwan 2,901, Netherlands 2,654, Italy 2,599,
  Australia 1,998
• Cellphones around 1 billion
• China 400 m, US 200 m, India 176 m (2010, 500 m)

92
Wireless Connectivity

• Wireless has accelerated electronic connectivity
• Easier deployment and maintenance
• Examples
• IEEE 802.11 / WiFi
• Cellular Networks
• Mobile Ad Hoc Networks
• Sensor Networks
• Mesh Networks / IEEE 802.16 / WiMAX