Embed Size (px)
Citation preview
Lero (c) 2014
Can Road Traffic Volume Information Improve Partitioning
for Distributed SUMO?
Ulrich Dangel, Quentin Bragard, Patrick McDonagh, Anthony Ventresque and Liam Murphy
Lero (c) 2014
TRAFFIC
• Software Engineering issues within Intelligent Transportation Systems
• VANET (NS3 & SUMO)
• Global Route balancing (SUMO)
• Distributed Simulation (SUMO)
Lero (c) 2014
Why SUMO?
• TAPAS, rl-world scenario, not 100x100 grid
• dSUMO
• TraCI
Lero (c) 2014
Objective
• Use traffic information to improve map partitioning
• Unbalanced partitions cause delay/waiting time
Lero (c) 2014
Datasets / Open Data
• Dublin - SCATS & TRIPS via dublinked
• SCATS preliminary data
• TRIPS - realtime
Lero (c) 2014
Dublin SCATS
Lero (c) 2014
Dublin TRIPS
Lero (c) 2014
Dublin TRIPS
Lero (c) 2014
Datasets / Open Data
• Dublin - SCATS & Trips via dublinked
• SCATS preliminary data
• TRIPS - realtime
• Census data
• Manual vehicle count
Scenario for inner city Dublin by Ronan Doolan
Lero (c) 2014
Partitioning
} map & reduce
Lero (c) 2014
Map Partitioning
Lero (c) 2014
Quadtree
Lero (c) 2014
Quadtree
Lero (c) 2014
Quadtree
0
5
10
15
20
Population (in Million)
North Rhine-Westphalia BavariaBaden-Württemberg Lower Saxony
Lero (c) 2014
Region Growing
Lero (c) 2014
Region Growing
Lero (c) 2014
Region Growing
Lero (c) 2014
Smart Quadtree
Lero (c) 2014
Smart Quadtree
Lero (c) 2014
Smart Quadtree
Lero (c) 2014
Region Growing
Lero (c) 2014
SParTSim
Lero (c) 2014
SParTSim
Lero (c) 2014
SParTSim
Lero (c) 2014
SParTSim
Lero (c) 2014
Heuristic
Lero (c) 2014
Heuristic
H(e) = dist(e)lanes(e)
Lero (c) 2014
Usage Data
Lero (c) 2014
Change Heuristic
• Use real-data / runtime data to improve partitioning
• Use volume data - TAPASCologne
• Can be calculated offline
Lero (c) 2014
Usage Data
Nw Node weight
ct total number of carswt
=ctn
cmax
ctn cars at node
Nw =X
wtctnct
Lero (c) 2014
Extensions
Lero (c) 2014
Extension Quadtree
Wp =X
Nwp
Lero (c) 2014
Extension Quadtree
Lero (c) 2014
Extension Smart Quadtree
XNwppartition to join =
Lero (c) 2014
Extension Smart Quadtree
Lero (c) 2014
Extension SParTSim
• Only start node selection was modified
• Trading phase uses the existing, established method
• Identify impact of starting node
Lero (c) 2014
Extension SParTSim
Lero (c) 2014
Equality
Lero (c) 2014
Equality QuadTree - 4
0.4
0.5
0.6
0.7
0.8
0 2000 4000 6000Step
Sim
pson
Inde
x
4 partitions
Lero (c) 2014
Equality QuadTree - 10
0.3
0.4
0.5
0.6
0 2000 4000 6000Step
Sim
pson
Inde
x
10 partitions − modified
10 partitions − unmodified
Lero (c) 2014
Equality - Smart QT - 4
0.4
0.6
0.8
1.0
0 2000 4000 6000Step
Sim
pson
Inde
x
4 partitions − modified
4 partitions − unmodified
Lero (c) 2014
Equality - Smart QT 8
0.3
0.4
0.5
0.6
0.7
0.8
0 2000 4000 6000Step
Sim
pson
Inde
x
8 partitions − modified
8 partitions − unmodified
Lero (c) 2014
Equality - SParTSim - 4
0.5
0.6
0.7
0.8
0 2000 4000 6000Step
Sim
pson
Inde
x
4 partitions − modified
4 partitions − unmodified
Lero (c) 2014
Equality - SParTSim 8
0.4
0.5
0.6
0.7
0 2000 4000 6000Step
Sim
pson
Inde
x
8 partitions − modified
8 partitions − unmodified
Lero (c) 2014
Conclusion
• Using volume data can improve partitioning
• Smart Quadtree (modified) can achieve better results than SParTSim
Questions?
Lero (c) 2014
Outlook
• Update dSUMO to support latest SUMO
• Switch to embedded Python
• Evaluate repartitioning strategies
Lero (c) 2014
Wishlist ;)
• Speed up TRaCI
• Maybe luas scripting interface?
Lero (c) 2014
Visualisation
Lero (c) 2014
Simpson Diversity Index
Dt =1P
p=0(Cp/Ct)2p
Lero (c) 2014
Communication - QT
0
500
1000
1500
0 2000 4000 6000Step
Mes
sage
s
4 partitions
10 partitions − modified
10 partitions − unmodified
Lero (c) 2014
Communication - Smart QT
0
1000
2000
3000
0 2000 4000 6000Step
Mes
sage
s 4 partitions − modified
4 partitions − unmodified
8 partitions − modified
8 partitions − unmodified
Lero (c) 2014
Communication
0
1000
2000
3000
4000
5000
0 2000 4000 6000Step
Mes
sage
s 4 partitions − modified
4 partitions − unmodified
8 partitions − modified
8 partitions − unmodified
