Fair Sharing of Bandwidth in VANETs
2nd Workshop on Vehicular Ad Hoc Networks
2nd September 2005, Cologne, Germany
Marc Torrent-Moreno, University of Karlsruhe, Germany
Paolo Santi, Consiglio Nazionale delle Ricerche, Italy
Hannes Hartenstein, University of Karlsruhe, Germany
VANET‘05
Marc Torrent-Moreno
Universität Karlsruhe (TH)
Which Problem are we looking at?
Simple scenario:
Simple question:
- Which transmission power should I use?
Not so simple constraint:
- Safety
VANET‘05
Marc Torrent-Moreno - 2
Universität Karlsruhe (TH)
Structure of this talk
»
Introduction
»
Motivation
»
The challenge
»
Our approach – Fairness
»
FPAV Algorithm
»
Experiments
»
Discussion – Current work
»
Summary
VANET‘05
Marc Torrent-Moreno - 3
Universität Karlsruhe (TH)
Introduction – Context
»
In a VANET every vehicle will send/receive into/from shared medium
»
In the USA, FCC: 7 channels, 1 control channel
»
Control channel: Exchange of safety information
»
2 types of safety related messages:
– Event driven
– Periodic
»
Periodic messages will essentially contain the state of the vehicle :
–
–
–
–
VANET‘05
Position
Direction
Speed
…
Marc Torrent-Moreno - 4
Universität Karlsruhe (TH)
Motivation
»
Amount of load resulting from beaconing should be limited
Why?
– Leave some bandwidth for unexpected emergency situations
– 802.11p is a variant of 802.11a with 10MHz channels (half data-rates)
– CSMA/CA is an asynchronous approach, not able to manage bandwidth
efficiently
– Safety messages are mainly broadcast (~Hidden terminal problem)
– Potential high node density scenarios
How?
– Prioritization – Admission control
– Adjusting packet generation rate
– Adjusting transmission power
VANET‘05
Marc Torrent-Moreno - 5
Universität Karlsruhe (TH)
The Challenge
»
Constraint: The load in the channel must be limited – MBL
(Maximum Beaconing Load)
»
… how do I make sure that the beaconing load is lower than
MBL everywhere in a reasonable way?
»
Options for power level assignments:
– Capacity
– Connectivity
VANETs main goal is Safety!
VANET‘05
Marc Torrent-Moreno - 6
Universität Karlsruhe (TH)
How do we achieve safety?
»
… being fair:
»
How can we adjust transmission power in a fair approach?
Fairness from the system perspective
Fair Power Control: All vehicles in a certain (busy) area must
restrict their beacons’ transmission power by the same ratio
– Maximize the minimum transmission power
– Maximize individual coverage
VANET‘05
Marc Torrent-Moreno - 7
Universität Karlsruhe (TH)
Our Approach to Fairness – Assumptions
»
Assumptions:
– The Carrier Sense Range is monotonic
Also in the paper for clarity reasons:
– All nodes have the same beaconing frequency and packet sizes
– All nodes have the same Maximum Transmission Power (Pmax)
»
In this case, MBL can be counted in:
– bits per second
– # of overlapping Carrier Sense Ranges
BL=2
VANET‘05
BL=3
Marc Torrent-Moreno - 8
Universität Karlsruhe (TH)
Our Approach to Fairness – Definition
»
Problem Statement:
BL(PA) ≤ MBL
BL=2
PA(1)
»
• PA: Power Assignment
• PA(i)×Pmax: Transmission Power
where PA(i)∈[0,1]
BL=3
PA(2)
PA(3)
PA(n)
Challenge:
– Beaconing Max-Min Tx Power
– Per node maximal power assignment
VANET‘05
Marc Torrent-Moreno - 9
Universität Karlsruhe (TH)
FPAV Algorithm
»
FPAV: Fair Power Adjustment for Vehicular environments
»
First approach – Optimal Solution:
– Totally centralized
– Global knowledge
»
Two stages:
– Stage 1: Compute an optimal solution to BMMTxP
– Stage 2: Augment the solution into a per-node maximal power
assignment
VANET‘05
Marc Torrent-Moreno - 10
Universität Karlsruhe (TH)
FPAV – Stage 1
Input:
a set of nodes
N = {u1 ,..., un }
Output: a power assignment PA which is an ( ε×Pmax-aproximation of an)
optimal solution to BMMTxP
∀ui ∈ N , set PA(i ) = 0
while ( BL( PA) ≤ MBL) do
∀ui ∈ N , PA(i ) = PA(i ) + ε
end while
∀ui ∈ N , PA(i ) = PA(i ) − ε
VANET‘05
Marc Torrent-Moreno - 11
Universität Karlsruhe (TH)
FPAV – Stage 1 – Visualization
BL
MBL = 2
2
1
VANET‘05
Marc Torrent-Moreno - 12
Universität Karlsruhe (TH)
FPAV – Stage 2
Input:
an optimal solution to BMMTxP, denoted PA
Output: a power assignment PA which is a ( ε×Pmax-aproximation of a)
per node maximal power assignment following strict fairness
∀ui ∈ N , set ReachedTop(i) = false
while !(∀ui ∈ N , set ReachedTop = true) do
for(i = 1 to n) do
PA(i) = PA(i) + ε
if ( BL( PA) > MBL) then
PA(i) = PA(i) − ε
ReachedTop(i) = true
end if
end for
end while
VANET‘05
Marc Torrent-Moreno - 13
Universität Karlsruhe (TH)
FPAV – Stage 2 – Visualization
BL
MBL = 2
2
1
Please, take a look at the paper for the proofs
VANET‘05
Marc Torrent-Moreno - 14
Universität Karlsruhe (TH)
Experiments
Configuration
Distribution
10 pckts/s
Packet size
250B
Loadvehicle
20Kbps
Data Rate
6Mbps
Maximum beaconing load 3Mbps
Communication range
250m
Vehicle Density (Vehicles/1000m)
250
Packet generation rate
Random Vehicle Density
Deterministic Vehicle Density
200
150
100
50
0
0
Carrier Sense range
500m
500
1000
1500
2000
2500
Position in the road
3000
3500
4000
#Cars
Step size ε
0.01
Vehicle density
20veh./100m
4
1
m
VANET‘05
Marc Torrent-Moreno - 15
Universität Karlsruhe (TH)
Experiments – Results
Deterministic Distribution
5000
1.8
1.6
1.2
1
0.8
2000
1.8
1.6
4000
1.4
3000
2
Offered Load
MaxBeaconingLoad
Adjusted Load
PA value
5000
Load (Kbps)
PA
Load (Kbps)
4000
2
1.4
1.2
3000
1
0.8
2000
PA values
0.6
0.4
1000
PA values
0.6
0.4
1000
0.2
0
0
500
1000
1500
2000
2500
Position in the road
VANET‘05
3000
3500
0
4000
0.2
0
0
500
1000
1500 2000 2500
Position in the road
3000
3500
0
4000
Marc Torrent-Moreno - 16
Universität Karlsruhe (TH)
PA
Offered Load
MaxBeaconingLoad
Adjusted Load
PA value
‘Random’ Distribution
Discussion – Current Work
»
Centralized or synchronous approaches are not feasible in VANETs
»
Challenge: iterative (feedback) schemes are likely to fail due to high
dynamics
Current Work:
»
How to implement FPAV in a fully distributed, asynchronous fashion
– Let each node run FPAV internally with local knowledge
– How close to centralized FPAV can we get?
VANET‘05
Marc Torrent-Moreno - 17
Universität Karlsruhe (TH)
Summary
»
We addressed the limited channel capacity of VANETs to support
safety-related information exchange
»
We propose to limit the wireless channel load resulting from periodic
messages
»
We propose strict FAIRNESS due to SAFETY nature of VANETs
»
Formally defined the challenge in terms of Max-Min optimization
problem extending it to obtain per-node maximality
»
We proposed a centralized algorithm FPAV and briefly point the
issues to bring it into a real scenario
»
Our goal is to come up with a fully distributed, localized and
asynchronous implementation
VANET‘05
Marc Torrent-Moreno - 18
Universität Karlsruhe (TH)
Thank you very much for your attention !
http://dsn.tm.uka.de
VANET‘05
Marc Torrent-Moreno
Universität Karlsruhe (TH)