Practical 3D Geographic Routing for Wireless Sensor Networks - - PowerPoint PPT Presentation

▶
practical 3d geographic routing for wireless sensor
SMART_READER_LITE
LIVE PREVIEW

Practical 3D Geographic Routing for Wireless Sensor Networks - - PowerPoint PPT Presentation

Oct 10, 2023 420 likes •1.45k views

Practical 3D Geographic Routing for Wireless Sensor Networks Jiangwei Zhou*, Yu Chen + , Ben Leong , Pratibha Sundar Sundaramoorthy * Xian Jiaotong University + Duke University National University of Singapore Geographic Routing

slide-1
SLIDE 1

Practical 3D Geographic Routing for Wireless Sensor Networks

Jiangwei Zhou*, Yu Chen+, Ben Leong∇, Pratibha Sundar Sundaramoorthy∇

*Xi’an Jiaotong University

+Duke University ∇National University of Singapore

slide-2
SLIDE 2

Geographic Routing Algorithms Exploit geometric information (coordinates) of network topology to improve scalability

  • f point-to-point routing
slide-3
SLIDE 3

Geographic Routing Algorithms Greedy forwarding + Recovery mode when local minimum is encountered

slide-4
SLIDE 4

1.Efficient 2.Storage proportional to density, not size

slide-5
SLIDE 5

Motivation

Previously proposed geographic routing algorithms assume “planar” network topology ⇒ Many modern sensor networks are three-dimensional

slide-6
SLIDE 6

Two Questions

  • 1. How do we get

geographic routing to work for 3D networks?

slide-7
SLIDE 7

Two Questions

  • 2. How do existing point-

to-point algorithms compare? (should we care?)

slide-8
SLIDE 8

Outline

  • Problem & Motivation
  • Overview of related work &

geographic routing

  • Our Solution: GDSTR-3D
  • Performance Evaluation
  • Conclusion
  • Future Work
slide-9
SLIDE 9

Related work

  • 2D geographic routing

― GPSR (Karp & Kung, Mobicom 2000) ― GOAFR+ family (Kuhn et al., Mobihoc 2003) ― CLDP (Kim et al., NSDI 2005) ― GDSTR (Leong et al., NSDI 2006)

  • 3D geographic routing

― GRG (Flury & Wattenhofer, Infocom 2008) ― GHG (Liu & Wu, Infocom 2009)

slide-10
SLIDE 10

Related work

  • Point-to-point

― AODV (Perkins, Milcom 1997) ― VPCR (Newsome & Song, SenSys 2003) ― BVR (Fonseca et al., NSDI 2005) ― VRR (Caesar et al., SIGCOMM 2006) ― S4 (Mao et al., NSDI 2007)

  • Virtual Coordinates

― NoGeo (Rao et al., Mobicom 2003) ― PSVC (Zhou et al., ICNP 2010)

slide-11
SLIDE 11

Our Approach

Extend GDSTR to 3D

Complications!

slide-12
SLIDE 12

Overview: Geographic Routing

Nodes have coordinates

slide-13
SLIDE 13

Overview: Geographic Routing

source

Nodes have coordinates

slide-14
SLIDE 14

Overview: Geographic Routing

source dest

Nodes have coordinates

slide-15
SLIDE 15

Overview: Geographic Routing

source dest

Packet contains coordinates of destination

slide-16
SLIDE 16

Overview: Geographic Routing

source dest

Greedy forwarding!

slide-17
SLIDE 17

Overview: Geographic Routing

source dest

Greedy forwarding!

slide-18
SLIDE 18

Overview: Geographic Routing

source dest

Greedy forwarding!

slide-19
SLIDE 19

Overview: Geographic Routing

source dest

Dead end! (local minima)

slide-20
SLIDE 20

Overview: GDSTR

source dest

Distributed Spanning Tree

slide-21
SLIDE 21

Overview: GDSTR

source dest

Distributed Spanning Tree

root

slide-22
SLIDE 22

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-23
SLIDE 23

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-24
SLIDE 24

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-25
SLIDE 25

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-26
SLIDE 26

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-27
SLIDE 27

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-28
SLIDE 28

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-29
SLIDE 29

Overview: GDSTR

Aggregate coordinates with convex hulls

root

slide-30
SLIDE 30

Overview: GDSTR

Hull Tree

root

slide-31
SLIDE 31

Overview: GDSTR

source dest

Remember minimum ⇒ tree traversal

root

slide-32
SLIDE 32

Overview: GDSTR

source dest

Tree Traversal

root

slide-33
SLIDE 33

Overview: GDSTR

source dest

Tree Traversal

root

slide-34
SLIDE 34

Overview: GDSTR

source dest

Back to Greedy Forwarding!

slide-35
SLIDE 35

Overview: GDSTR

source dest

Back to Greedy Forwarding!

slide-36
SLIDE 36

Overview: GDSTR

source dest

Done!!

slide-37
SLIDE 37

Why do hull trees work?

  • Used only to escape from

local minimum

  • Cheap to build – O(log n)
slide-38
SLIDE 38

Caveat

CONCAVE VOID

slide-39
SLIDE 39

Caveat

dest

slide-40
SLIDE 40

Caveat

dest local minimum

slide-41
SLIDE 41

Caveat

dest root

TERRIBLE!

local minimum

slide-42
SLIDE 42

Need TWO hull trees rooted at opposite ends

dest root local minimum

One tree sufficient for correctness. Two trees needed for efficiency.

slide-43
SLIDE 43

Our Approach

Extend GDSTR to 3D

Complications!

slide-44
SLIDE 44

Challenges

(Why is it hard in TinyOS?)

  • TinyOS does not support dynamic

memory allocation

  • CC2420 radio supports up to 128

bytes in size and has a limited data rate

  • Limited DRAM and flash memory
  • Precision of floating point operations is

limited

slide-45
SLIDE 45

Naïve Implementation of 3D Convex Hull

  • Computations are costly
  • Need to store auxiliary data

structures for efficiency ⇒ storage costly

  • Messages too big
slide-46
SLIDE 46

Key ideas

  • 1. Approximate 3D Convex

Hull with 2 x 2D Convex Hull

  • 2. Use two-hop greedy

forwarding

  • 3. Simplify (details in paper)
slide-47
SLIDE 47

GDSTR-3D

x y z

Example of a 3D convex hull

slide-48
SLIDE 48

GDSTR-3D

x y z

Projection onto

  • rthogonal planes

(xy-, yz-, and zx-plane)

slide-49
SLIDE 49

GDSTR-3D

x y z

Projection onto

  • rthogonal planes

(xy-, yz-, and zx-plane)

slide-50
SLIDE 50

GDSTR-3D

x y z

Use two of these 2D convex hulls to approximate the 3D convex hull

slide-51
SLIDE 51

PERFORMANCE EVALUATION

slide-52
SLIDE 52

Metrics

  • 1. Success rate
  • 2. Hop stretch
  • 3. Maximum Storage
  • 4. Message Overhead
slide-53
SLIDE 53

Indriya Testbed (NUS)

  • 127 TelosB motes

distributed over 3 floors

  • Picked random subsets
  • f nodes on 1, 2 and 3

floors

slide-54
SLIDE 54

Algorithms

  • 1. GDSTR-3D
  • 2. GDSTR
  • 3. CLDP/GPSR
  • 4. AODV
  • 5. VRR
  • 6. S4

(2D Face Routing)

(-2D)

slide-55
SLIDE 55

Success rate vs. network size

slide-56
SLIDE 56

Hop stretch (GDSTR+) vs. network size

One-hop Two- hop

slide-57
SLIDE 57

Hop stretch vs. network size

slide-58
SLIDE 58

Size of compiled binaries & source code

Algorithm Compiled binary Size (KB) Lines of code

GDSTR-3D 39.5 2,757 GDSTR 33.8 2,641 CLDP/GPSR 47.5 2,500 S4 43.2 3,997 VRR 45.1 4,135 AODV 21.1 1,294

slide-59
SLIDE 59

TOSSIM Experiments

slide-60
SLIDE 60

Hop stretch vs. network density

2D doesn’t work!

slide-61
SLIDE 61

Greedy forwarding success rate

  • vs. network density
slide-62
SLIDE 62

Scaling Up

slide-63
SLIDE 63

Hop stretch vs. network size

slide-64
SLIDE 64

Maximum storage vs. network size

slide-65
SLIDE 65

Message overhead (bytes) vs. network size

slide-66
SLIDE 66

Algorithm Stretch Storage Overhead

GDSTR-3D GDSTR-2D S4 VRR AODV

  • ?

Summary: Scaling Up (3,200 nodes)

slide-67
SLIDE 67

Comprehensive comparison

  • f GDSTR-3D to

1.AODV 2.VRR 3.S4 Details in the paper.

slide-68
SLIDE 68

Key Contributions

  • 1. Practical 3D geographic

routing

  • 2x2D hulls for aggregation
  • Two-hop greedy
  • 2. Comprehensive comparison
  • f state-of-art point-to-point

algorithms for TinyOS

slide-69
SLIDE 69

Summary

For small sensor networks (<200 nodes): pick your favorite

  • algorithm. 
slide-70
SLIDE 70

For large sensor networks (~3,200 nodes), geographic routing algorithms are most scalable:

  • relatively low overheads
  • storage matters, but is not
  • verriding consideration
slide-71
SLIDE 71

Life’s complicated

slide-72
SLIDE 72

Algorithm Needs coordinates? Needs location service Reactive?

GDSTR-3D S4 VRR AODV

Tradeoffs at a glance

slide-73
SLIDE 73

Future Work

  • More Thorough Comparison
  • link losses
  • quantify cost of location service/

coordinate assignment

  • resilience
  • incremental costs
  • traffic pattern/load
  • Sleep-wake duty cycle
  • Reduce memory footprint
slide-74
SLIDE 74

TinyOS Source Code

Available here: https://sites.google.com/site/geographicrouting Or email me: benleong@gmail.com

slide-75
SLIDE 75

Questions?

slide-76
SLIDE 76

Thank You

slide-77
SLIDE 77

For large sensor networks , geographic routing algorithms are most scalable:

  • guarantee packet delivery
  • storage cost is proportional to network

density but size

  • motes have small RAM
slide-78
SLIDE 78

Choice:

  • Extend existing 2D geographic routing

algorithms to implement a 3D routing algorithm

  • GDSTR is a natural candidate for

extension

slide-79
SLIDE 79

Routing in 3D:

  • Geographic routing in 3D topologies is intrinsically

harder than routing in 2D topologies since greedy forwarding tends to encounter more local minima in general 3D topologies

  • It is not entirely straightforward to extend GDSTR to

3D because that 3D convex hulls require significantly more storage and are much more computationally costly

slide-80
SLIDE 80

Solution:

  • Extend greedy forwarding by using 2-

hop neighbor information to improve the greedy forwarding success rate in 3D networks

  • Approximate 3D convex hulls with two

2D convex hulls

slide-81
SLIDE 81

All graphs

slide-82
SLIDE 82

Greedy forwarding success rate

  • vs. network size(high density)
slide-83
SLIDE 83

Greedy forwarding success rate

  • vs. network size(low density)
slide-84
SLIDE 84

Success rate vs. network size

slide-85
SLIDE 85

Hop stretch(GDSTR+) vs. network size

slide-86
SLIDE 86

Hop stretch vs. network size

slide-87
SLIDE 87

Greedy forwarding success rate

  • vs. network density
slide-88
SLIDE 88

Hop stretch vs. network density

slide-89
SLIDE 89

Average storage vs. network density

slide-90
SLIDE 90

Maximum storage vs. network density

slide-91
SLIDE 91

Message overhead(packets)

  • vs. network density
slide-92
SLIDE 92

Message overhead(bytes)

  • vs. network density
slide-93
SLIDE 93

Scaling up

slide-94
SLIDE 94

Greedy forwarding success rate

  • vs. network size(low density)
slide-95
SLIDE 95

Greedy forwarding success rate

  • vs. network size(high density)
slide-96
SLIDE 96

Hop stretch vs. network size(low density)

slide-97
SLIDE 97

Hop stretch vs. network size(high density)

slide-98
SLIDE 98

Hop stretch vs. network size with multiple obstacles

slide-99
SLIDE 99

Hop stretch vs. network size with multiple obstacles

slide-100
SLIDE 100

Maximum storage vs. network size

slide-101
SLIDE 101

Message overhead(bytes) vs. network size