+ Network Research Lab, University of CaliforniaLos Angeles, California, USA
* University of Karlsruhe in the Karlsruhe Institute of TechnologyKarlsruhe, Germany
GeoDTN+Nav: A Hybrid Geographic and DTN Routing with Navigation Assistance in Urban Vehicular Networks
Pei-Chun Cheng+, Jui-Ting Weng+, Lung-Chih Tung+, Kevin C. Lee+, Mario Gerla+, Jérôme Härri*
1st Annual Symposium on Vehicular Computing Systems (ISVCS)
Dublin, IrelandJuly 23rd , 2008
2
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Outline
� Background and Motivation
� Navigation-Assist Model – VNI
� GeoDTN+Nav
� Evaluation
� Conclusion and Future Work
3
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Motivation
� Geographic Routing:
� routes packets based on geographic information of the source, relays and
destination
� Based on a greedy forwarding approach
� Efficient in this situation:
� and in this one ??
s d
d
s ?
4
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Motivation
� A second phase called “recovery” exists..
�And in this one ??
�Need to use nodes mobility to move between partitions
� Low density
� Spatial and temporal mobility patterns
� Obstacles
d
s
d
s
d
s
5
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Goal of GeoDTN+Nav
� Aim to improve packet delivery in a disconnected VANET
� Targeted for DTN
� If delay is not crucial
� Inspired from
� LER (Last Encounter Routing),
� ZebraNet,
� Data mules…
� Basic idea
� Exploit vehicles’ mobility
� Use store-carry-forward instead of store-forward
� Challenge:
� Finding the correct “Mule”
6
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Outline
� Background and Motivation
� Navigation-Assist Model – VNI
� GeoDTN+Nav
� Evaluation
� Conclusion and Future Work
7
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Navigation-assisted Model - VNI
� Basic idea is trivial
� Exploit mobility to help deliver packets across disconnected networks
� The problem now is which neighbor to choose?
� Blind random choice
� Might not help
� Neighbors may move even farther away from the destination
� “Informed” choice
� Better decision
� If we know neighbors’ destination or path information
� How to know such Information ?
8
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Navigation Systems
� General Idea� Exploit navigation systems
� Harvest neighbors’ destination/path information
� Assumption� Every vehicle has a navigation system
� Navigation system becomes common accessory� Provide real-time traffic information
� Not always true !
� Relaxed Assumption� “Pseudo/Virtual” navigation system
9
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Categorizing Vehicular Mobility Patterns
Categories Example
Deterministic(Fixed) RouteBig Blue Bus, UCLA Shuttle,
Metro Train
Deterministic(Fixed) Destination Taxi, UCLA Van pool
Probabilistic(Expected) Route /
DestinationNavigation system guided vehicles
Unknown Random movement
� We propose a “Virtual Navigation Interface” to generalize these
vehicle categories
10
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Virtual Navigation Interface - VNI
� Assume VNI is installed on every vehicle
� A lightweight wrapper interfaces: interact with data sources
� Provide two unified information:
� Route info
� Destination
� Path
� Direction
� Confidence
� 0% (Random) ~ 100%(Deterministic)
11
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Virtual Navigation Interface - VNI
w/ Navigation
VNI: (path, 25%)
w/o Navigation
VNI: (?, 0%)
Bus
VNI: (trip, 100%)
Taxi
VNI: (dest, 100%)
12
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Outline
� Background and Motivation
� Navigation-Assist Model – VNI
� GeoDTN+Nav
� Evaluation
� Conclusion and Future Work
13
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
� Introduce a third forwarding mode in georouting
� DTN recovery mode
� Complement conventional two-mode geo-routing
� Three routing modes
� Greedy
� Perimeter
� DTN
14
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Algorithm
� Every node adds the VNI tuple to the periodic beacons
� “Route_info, Confidence”
� Follow conventional Greedy mode as long as possible
� In Perimeter forwarding
� Detect possible partitions
� Check neighbors’ navigation information
� Switch to DTN mode if necessary
� Network partitioned
� “mule” may bring the packet out of the local maximum
15
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
When to Switch to DTN mode ?
� Computes a switch score in perimeter mode:
� S(i)= αP(h) • βQ(i) • γDir(i)
� Where:
� P(h): partition probability
� Q(i): Quality of the “mule”
� Dir(i): Direction of the “mule”
� If S(i) > Sthres => Switch to DTN mode
16
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
� Let
� α = β = γ = 1
� Sthres = 0.25
Example
Q(N1) = 0.1
S(N1) = P(8) * Q(N1)
= 0.04
P(9) = 0.5
Q(N2) = 0
S(N2) = 0
P(8) = 0.4
Q(N3) = 0.6
S(N3) = 0.24
Q(N1) = 0.2
S(N1) = 0.01
Q(N2) = 0.7
S(N2) = 0.35 Q(N3) = 0.6
S(N3) = 0.30
17
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Outline
� Background and Motivation
� Navigation-Assist Model – VNI
� GeoDTN+Nav
� Evaluation
� Conclusion and Future Work
18
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Sythetic Scenario
� Qualnet 3.95, 802.11b, transmission rate 2Mbps, radio range 350m
� α = β = γ = 1, Sthres = 0.25
� Static nodes randomly placed
� Mules moving at 50km/hr, varies from 5 to 40 mules
� Inter-departure time at A and B: � Uniform
� Random
� Fixed source and destination
� 20 runs
� Compare GPCR and GeoDTN+Nav for� PDR
� latency
19
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Realistic Scenario
� Oakland, 1500m by 4000m
� Mobility generated by VanetMobisim
� Additional mules generated (5 to 40)
� Random source and fixed destination
� 20 runs each
� Compare GPSR, GPCR, and GeoDTN+Nav
� PDR
� latency
20
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Synthetic Scenario (PDR)
� GeoDTN+Nav has a PDR
70% PDR above GPCR in
both inter-departure
� Almost Full connectivity
explains the sharp increase in
PDR for GPCR/UNIM
21
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Synthetic Scenario (Latency)
� GeoDTN+Nav’s latency drops
as more mobile nodes
increase connectivity
22
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Realistic Scenario (PDR)
� GeoDTN+Nav has higher PDR
than GPSR and GPCR in both
random and uniform inter-
departure time
23
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Realistic Scenario (Latency)
� GPCR’s latency remains small
as only 1-hop connectivity
24
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Outline
� Background and Motivation
� Navigation-Assist Model – VNI
� GeoDTN+Nav
� Evaluation
� Conclusion and Future Work
25
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Conclusion
� We proposed a generalized framework for navigation information � Different level of information for privacy: Path, Destination, Direction
� Consider uncertainty: Confidence
� We presented a hybrid georouting protocol with DTN mode� Use DTN to when network is partitioned and georouting fails
� For DT applications
� Use VNI to select the appropriate “mule” for DTN mode
� Showed it was able to deliver packets in partitioned and sparse networks
� Future work:� Increase the mules heterogeneity
� Parameter optimization: setting switch score threshold adaptively based on topology connectivity
� Cache policy: How much to cache? What is the right buffer capacity?
26
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Synthetic Scenario (Hop Count)
� GeoDTN+Nav has higher hop
count due to higher PDR
� Closeness in PDR explains
the hop count convergence of
GPCR/UNIM and
GeoDTN+Nav/UNIM from 25
onward
27
ISVCS 2008
Jérôme Härri
Traffic Telematics –
DSN Research Group – Institute of Telematics
Evaluation – Realistic Scenario (Hop Count)
� GeoDTN+Nav has higher hop
count due to higher PDR
� 1 hop for GPSR and GPCR
shows that there is NO
connectivity except src/dest
pair that is one hop away!!!
� GeoDTN+Nav operates under
intermittent connectivity!