Data Preference Matters:
A New Perspective of Safety Data Dissemination
in Vehicular Ad Hoc Networks
Qiao Xiang1 , Xi Chen1 , Linghe Kong1 , Lei Rao2 and Xue Liu1
1 School
of Computer Science, McGill University
Motors Research Lab
2 General
April 29th, 2015
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
1/ 31
Introduction
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
2/ 31
Introduction
Background
Introduction
Vehicular Ad Hoc Networks
Communication infrastructure for Intelligent
Transportation Systems (ITS)
Operate in a dynamic environment
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
3/ 31
Introduction
Background
Introduction
Safety Data Dissemination
Crucial for vehicle safety
Contain periodic routine data and event-driven emergency
data
More emphasis on QoS, e.g., small delay and high
coverage
(a) Collision Avoidance
(b) Lane Change
Sources:www.gm.com and www.Mercedes-Benz.com
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
4/ 31
Introduction
Background
Data Preferences
When collecting safety data, vehicles have
preferences on
Closer, Newer and More important data
Related works do not consider these preferences
Counter-Based Dissemination
Farthest-First Dissemination
Probabilistic Forwarding
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
5/ 31
Introduction
Our Focus
Our Focus
Quantifying data preferences
Designing lightweight distributed dissemination
protocol
Satisfying data preferences of all the vehicles
Understanding system benefits
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
6/ 31
Introduction
Our Focus
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
7/ 31
Quantifying Data Preferences
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
8/ 31
Quantifying Data Preferences
Data Preferences
Vehicles show the following preferences when
collecting safety data:
Spatial preference: the closer, the better;
Temporal preference: the newer, the better;
Type preference: the more important, the better.
Quantify these preferences on a per-packet level
Packet Value = Spatial Value×Temporal Value×Type Value.
Given a packet p, its packet-value for vehicle v :
PVv (p) = Sv (p) · Tv (p) · Wp .
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
9/ 31
Quantifying Data Preferences
Spatial-Value Function
Given a data packet p, its spatial-value
decreases as p is disseminated away from its
origin;
becomes zeros after p exceeds the Range of
Interest.
Figure: Region of Interest
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
10/ 31
Quantifying Data Preferences
Spatial-Value Function
Sv(p)
Range of Interest
0
Distance
Figure: Spatial-Value vs. Dissemination Distance.
Sv (p) =
max(α − βdpv , 0), if vehicle v moves towards (xp , yp )
0
otherwise.
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
11/ 31
Quantifying Data Preferences
Temporal-Value Function
Collecting real-time data is crucial for safety
applications;
New packets have much higher temporal-value
than old ones;
Figure: New data should be disseminated first.
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
12/ 31
Quantifying Data Preferences
Temporal-Value Function
Temporal-value decreases as time elapses;
The decreasing speed becomes slower with time;
Old packets still have value, e.g., for statistic
analysis.
Tv(p)
Tv (p) = e −µtypep te (p) .
Emergency Message
Routine Message
0
Elapsed Time
Figure: Temporal-Value vs. Elapsed Time.
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
13/ 31
Quantifying Data Preferences
Type-Value Function
Emergency messages are more important than
routine messages.
Wp
Emergency Message
Routine Message
0
Figure: Type-Value vs. Message Types.
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
14/ 31
Quantifying Data Preferences
Quantifying Data Preferences
Putting pieces together,
PVv (p) = Sv (p) · Tv (p) · Wp .
Packet-value data preference: Given any two
packets p1 and p2 , vehicle v always has a higher
data preference to p1 over p2 if PVv (p1 ) > PVv (p2 ).
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
15/ 31
PVCast: a Packet-Value-Based Dissemination Protocol
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
16/ 31
PVCast: a Packet-Value-Based Dissemination Protocol
PVCast in A Nutshell
PVCast in A Nutshell
A new packet p
Packet Value
Update
No
PVA(p)=0
1-Hop Dissemination Utility
Computation
Probabilistic Broadcast Test
Fail
Discard packet
Contention Window Size
Assignment
Broadcast
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
17/ 31
PVCast: a Packet-Value-Based Dissemination Protocol
1-Hop Dissemination Utility
1-Hop Dissemination Utility
Utility = Packet Value × Effective Dissemination Coverage.
Effective Dissemination Coverage
U
A
edcA (p) = 5 − 2 = 3
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
18/ 31
PVCast: a Packet-Value-Based Dissemination Protocol
Probabilistic Broadcast Test
Probabilistic Broadcast Test
Broadcast Probability
Piecewise function of dissemination utility
Higher dissemination utility
→ Higher chance for broadcasting
0
Qiao Xiang et al. (McGill)
Dissemintation Utility
IEEE INFOCOM’15
04/29/2015
19/ 31
PVCast: a Packet-Value-Based Dissemination Protocol
Contention Window Size Assignment
Contention Window Size Assignment
Minimum CW Size
Piecewise function of dissemination utility
Higher dissemination utility
→ Smaller minimum CW size
→ Higher priority to get channel access
0
Qiao Xiang et al. (McGill)
Dissemintation Utility
IEEE INFOCOM’15
04/29/2015
20/ 31
Performance Evaluation
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
21/ 31
Performance Evaluation
Simulation Settings
Simulation Settings
2000m × 30m two-way road
Speed limit: 100kph
SUMO-generated vehicle trace,
N ∈ {20, 40, 60, 80, 100}
Transmission range: 300m
Range of interest: 1200m
Routine message: 10/sec
Emergency message: 1/sec with 0.5 probability
in certain segment
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
22/ 31
Performance Evaluation
Simulation Settings
Simulation Settings
Comparison
CBD, FARTHEST, slottedP and PVCast
Metrics
Per-Vehicle Throughput
Broadcast Rate
Broadcast Efficiency
Per-Packet Delivery Delay
Per-Packet Vehicle Coverage
Per-Vehicle Emergency Throughput
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
23/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Per-Vehicle Throughput (pkt/sec)
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
24/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Broadcast Rate
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
25/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Broadcast Efficiency
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
26/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Per-Packet Delivery Delay (ms)
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
27/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Per-Packet Vehicle Coverage
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
28/ 31
Performance Evaluation
Evaluation Results
Evaluation Results
Per-Vehicle Emergency Throughput (pkt/sec)
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
29/ 31
Conclusion and Future Work
Outline
1
Introduction
2
Quantifying Data Preferences
3
PVCast: a Packet-Value-Based Dissemination Protocol
4
Performance Evaluation
5
Conclusion and Future Work
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
30/ 31
Conclusion and Future Work
Conclusion and Future Work
Conclusion
Quantification of data preferences in VANET
Integration of data preferences in the design of
dissemination protocol
System benefits in terms of low latency and high
coverage
Future Work
A more comprehensive model of data preference
Joint adaption of power, data rate and
contention window based on packet-value
Qiao Xiang et al. (McGill)
IEEE INFOCOM’15
04/29/2015
31/ 31