Efficient Geographic Routing in Multihop Wireless Networks

1
Efficient Geographic Routingin Multihop Wireless
Networks

• Seungjoon Lee
• Department of Computer Science,
• University of Maryland, College Park
• Joint work with Bobby Bhattacharjee, Suman
  Banerjee
• MobiHoc05
• Presented by Ting-Yu Lin

2
Outline

• Motivation
• New Link Metric for Geographic Routing
• Definition of Normalized Advance (NADV)
• Example of NADV
• Optimality of NADV in Idealized Environments
• NADV with Various Types of Link Cost
• Link Cost Estimation
• Simulation Results
• Conclusions

3
Routing in Wireless Networks
Minimum-cost path
S
T
Shortest path

• Shortest path is not always the best.
• We want to find a minimum-cost path
• ETX (Expected Transmission Count), Delay,
  Transmission energy consumption

4
Cost-based Routing
S
T
Shortest-hop path
5
Cost-based Routing
S
T
Shortest-hop path cost12
2
4
3
3

• ETX, Delay, Transmission energy consumption

6
Cost-based Routing
Alternative path cost9
3
2
1
1
2
S
T
Shortest-hop path cost12
2
4
3
3

• ETX, Delay, Transmission energy consumption
• Relatively easy to incorporate in on-demand
  routing
• How to find low-cost paths using geographic
  routing?

7
Geographic Routing

• Uses location information
• Each node knows its location (e.g., using GPS).
• Sources know the locations of destinations.
• Neighbor location is known through periodic
  messages.
• Next-hop Decision
• Greedy Forwarding
• Recovery Operation
• To overcome local minima, or voids
• All operations are local (in contrast to DSR,
  etc.)

8
Greedy Geographic Forwarding
D(S)
S
T
n
D(n)
Maximize advance ADV(n) D(S) - D(n)
9
Greedy Geographic Forwarding
Advance ADV(n) D(S) - D(n)
S
T
n
Our goal gt Maximize advance vs. Use low-cost link
gt Trading-off proximity and link cost
10
Example of Greedy Forwarding
S
T
n
Area with high error rate

• ADV does not consider link quality.
• We want BOTH large advance AND low link cost.

11
Normalized Advance (NADV)

• Definition NADV(n) ADV(n)/Cost(n)
• Advance per unit cost
• as link cost, then
• (expected advance per transmission)
• Goal
• Use NADV as link metric in geographical routing,
  such that a node forwards packets to the neighbor
  with largest NADV.

12
Benefits of NADV

• Applicable to many cost types
• E.g., ETX, transmission energy consumption, link
  delay
• General framework
• Different routing strategies depending on
  objectives
• E.g., Min-latency path, min-energy path
• Opportunity for Adaptive Routing
• Due to local next-hop decision in geographic
  routing

detour
C
S
A
B
T
Cost increase
13
Example of NADV with Lossy Links

• Need to avoid nodes in the gray zone
• Link cost ETX

14
Example of NADV with Lossy Links

• Need to avoid nodes in the gray zone
• Link cost ETX

Best for NADV
Best for ADV
Sender
Destination
15
Path Optimality using NADV (theoretical base)

• Goal
• Minimize the sum of link costs on the path.
• Assumptions (Idealized Environment)
• We can find a node at an arbitrary point (high
  density).
• Link cost increasing convex function of
  distance.
• E.g., Required transmission power is larger for
  longer distance
• Optimal Strategy
• To choose nodes only on the straight line between
  S and T.
• To choose nodes on an equidistant basis (all
  links have the same distance ADVx optimal
  interval).
• To choose the neighbor with minimum Costx/ADVx,
  or maximum NADVxADVx/Costx

16
Proposed WISE (Wireless Integration Sublayer
Extension)

• Located on top of MAC.
• Closely coordinates with MAC for efficient link
  cost estimation.
• Provides simple primitives for upper-layer
  protocols.

17
NADV with Various Cost Types

• Link Error Rate
• ETX 1/(Packet Success Ratio)
• NADV ADV/ETX ADV (Packet Success Ratio)
• Seada04, Zorzi03
• Link Delay
• Transmission time due to different link
  transmission rate
• NADV ADV/(Transmission Time)
• Packet Transmission Power
• NADV ADV/Cpower

18
Link Cost Estimation

• NADV requires link cost estimation.
• E.g., Packet error rate estimation using SNR or
  probe messages
• Fast and accurate estimation with lower overhead
  is desirable
• PER estimation schemes
• Probe messages
• SNR
• where
  SNR
• Self monitoring
• Detailed estimation schemes are in the paper.
• Multiple schemes for various operating
  environments

19
Simulation Experiments (ns-2)

• Scenarios
• 100 stationary nodes randomly placed in
  1km-by-1km square (250m tx range)
• One pair of src-dst (900 meters) CBR traffic 1
  pkt every 2 seconds
• Multiple sources
• Mobile nodes
• Link cost types
• ETX
• Link cost is a function of distance
• Link delay
• Transmission energy consumption
• Random link cost
• Link cost is NOT a function of distance

20
Delivery Ratio vs. Link Error RateNADV Achieves
High Delivery Ratio.

• Link Metric ADV/ETX

Higher packet error rate
21
Simulation ResultRetransmissions vs. Link Error
Rate

• 100 stationary nodes in 1km-by-1km square (250m
  tx range)
• One pair of src-dst (900 meters) CBR traffic 1
  pkt every 2 seconds.
• Uniformly random assignment of link error
  probability 0max-PER.

22
Simulation Result (using link estimation
schemes)Delivery Ratio vs. Link Error Rate

• 100 stationary nodes in 1km-by-1km square (250m
  tx range)
• One pair of src-dst (900 meters) CBR traffic 1
  pkt every 2 seconds.
• Change in background noise values

Delivery Ratio
23
Random Link CostNADV Finds Good Paths.

• Assign link cost randomly between 1 and 6
• Not a monotonic function of distance
• Link Metric NADV ADV/(Random Link Cost)
• Modify AODV to find a minimum-cost path (AODV).
• To minimize the sum of all link costs on the path.

24
Random Link CostNADV Finds Good Paths.

• Assign link cost randomly between 1 and 6
• Not a monotonic function of distance
• Link Metric NADV ADV/(Random Link Cost)
• Modify AODV to find a minimum-cost path (AODV).
• To minimize the sum of all link cost on the path.

• NADV is comparable to AODV and optimal routing
• Even with local, greedy decision for the next
  hop.
• Even when link cost is not a function of distance.

25
Random Link CostNADV Finds Good Paths.

• Assign link cost randomly from 1,6
• Not a function of distance
• Modify AODV to find a minimum-cost path.

26
Simulation ResultComparison with AODV

• Assign link cost randomly from 1,6
• Not a function of distance
• Modify AODV to find a minimum-cost path.

Randomly choose 50 of links and increase the
cost by 50.
27
Simulation ResultExperiment with Link Cost
Change

• Generic cost
• Random variable in 16
• Cost change during the communication

detour
S
T
Cost increase
28
Related Work
Geographic Routing GPSR, face routing
Link Error DeCouto03
Link Delay Awerbuch04
Xmit Energy Banerjee02
Other Cost
29
Related Work
Seada04 Zorzi03
SP-PowerStoj01
Geographic Routing GPSR, face routing
Link Error DeCouto03
Link Delay Awerbuch04
Xmit Energy Banerjee02
Other Cost
30
Related Work
NADV (Normalized Advance)
Seada04 Zorzi03
SP-PowerStoj01
Geographic Routing GPSR, face routing
Link Error DeCouto03
Link Delay Awerbuch04
Xmit Energy Banerjee02
Other Cost
31
Conclusions

• Normalized Advance (NADV)
• General Link Metric for Geographic Routing
• Applicable to various cost types
• Optimal
• Requires cost estimation schemes
• Performance Improvement
• Up to 5 times higher delivery ratio
• Comparable to optimal routing
• Future works
• Hybrid use of link costs
• Experiments on a real testbed