Geographic Centroid Routing for Vehicular Networks Effects of GPS - - PowerPoint PPT Presentation
Geographic Centroid Routing for Vehicular Networks Effects of GPS Error on Geographic Routing Dr. Justin P. Rohrer jprohrer@nps.edu Naval Postgraduate School TaNCAD Lab ( https://tancad.net ) VEHICULAR June 25, 2018 Abstract Geographic
Effects of GPS Error on Geographic Routing
jprohrer@nps.edu
Naval Postgraduate School TaNCAD Lab (https://tancad.net)
VEHICULAR – June 25, 2018
Geographic Centroid Routing for Vehicular Networks
Paper Abstract [Rohrer, 2018]
A number of geolocation-based Delay Tolerant Networking (DTN) routing protocols have been shown to perform well in selected simulation and mobility scenarios. However, the suitability of these mechanisms for vehicular networks utilizing widely-available inexpensive Global Positioning System (GPS) hardware has not been evaluated. We propose a novel geolocation-based routing primitive (Centroid Routing) that is resilient to the measurement errors commonly present in low-cost GPS devices. Using this notion
their performance with respect to positional errors as well as traditional DTN routing metrics. We show that they outperform existing approaches by a significant margin.
www.nps.edu 2 / 39
Introduction Background Centroid-Based Routing Simulations Conclusion
www.nps.edu 3 / 39
A Word About Us...
Justin P. Rohrer:
◮ Assistant Professor, Network Group,
Computer Science Department, Naval Postgraduate School
◮ Teaches: CS3502 (Networks I), CS4538 (Wireless Security),
CS4558 (Network Traffic Analysis)
◮ Developed, teaches: CS4554 (Network Modeling and Analysis) ◮ Leads: Center for Tactical Networked Communications @NPS ◮ Pi/Co-PI on Disruption Tolerant Networking and Network
Measurement Projects
www.nps.edu 4 / 39
TaNCAD Lab @ NPS
Naval Postgraduate School (NPS)
◮ Navy’s Research University ◮ Located in Monterey, CA ◮ ≃1500 graduate students: military officers & DoD civilians
Center for Tactical Networked Communication Architecture
◮ 3 NPS professors, 2 NPS staff ◮ Sponsors: USN, USMC, NSF, NSA, ONR, DARPA, . . .
Focus:
◮ Network Survivability and Resilience ◮ Disruption Tolerant Networks for Tactical Environments
www.nps.edu 5 / 39
Goals for this work
◮ Quantify the potential effects of GPS positional error on
geolocation-based routing protocols
◮ Evaluate the effectiveness of a simple routing primitive
designed to mitigate those effects
www.nps.edu 6 / 39
Goals for this work
◮ Quantify the potential effects of GPS positional error on
geolocation-based routing protocols
◮ Evaluate the effectiveness of a simple routing primitive
designed to mitigate those effects
◮ Not “create the worlds best DTN routing protocol”
www.nps.edu 6 / 39
Goals for this work
◮ Quantify the potential effects of GPS positional error on
geolocation-based routing protocols
◮ Evaluate the effectiveness of a simple routing primitive
designed to mitigate those effects
◮ Not “create the worlds best DTN routing protocol”
◮ (That’s future work...) www.nps.edu 6 / 39
Introduction Background Centroid-Based Routing Simulations Conclusion
www.nps.edu 7 / 39
Disruption Tolerant Routing
Disruption tolerance is the ability of a system to tolerate disruptions in connectivity among its components. Disruption tolerance includes tolerance of environmental challenges, weak and episodic channel connectivity, mobility, long & unpredictable delay, energy and power constraints.
◮ Make local forwarding decisions without globally consistent
information
◮ More information = better decisions ◮ Geolocation is an increasingly common external data source
◮ Readily available www.nps.edu 8 / 39
GPS Positional Samples
◮ GPS hardware receiver constantly updates estimated position ◮ Software applications sample position periodically ◮ Per-sample precision and accuracy vary widely
◮ Receiver quality (cost), antenna gain, view of sky all factors ◮ Cheap receivers with tiny antennas and occluded view of sky
are typical
◮ How bad?
www.nps.edu 9 / 39
GPS Positional Samples
◮ GPS hardware receiver constantly updates estimated position ◮ Software applications sample position periodically ◮ Per-sample precision and accuracy vary widely
◮ Receiver quality (cost), antenna gain, view of sky all factors ◮ Cheap receivers with tiny antennas and occluded view of sky
are typical
◮ Some days it’s really bad...
www.nps.edu 9 / 39
GPS Positional Samples
◮ GPS hardware receiver constantly updates estimated position ◮ Software applications sample position periodically ◮ Per-sample precision and accuracy vary widely
◮ Receiver quality (cost), antenna gain, view of sky all factors ◮ Cheap receivers with tiny antennas and occluded view of sky
are typical
◮ Advertised (intentional) error is ±20 m ◮ Error can be order of 1000 m ◮ 100 m error commonplace [Ben-Moshe et al., 2011]
www.nps.edu 9 / 39
Commercial GPS navigation
◮ Commercial GPS navigation devices do significant
post-processing of samples, snap to nearest road, and present best-guess position to user
◮ Smartphones incorporate data from WiFi and cellular radios ◮ All these innovations driven by errors/limitations in using GPS
position only
www.nps.edu 10 / 39
Network Simulators
◮ Position typically provided as cartesian coordinates ◮ Location is accurate to arbitrary precision ◮ Geographic routing protocols are designed and tested under
these conditions (mobile testbeds are hard)
www.nps.edu 11 / 39
Network Simulators
◮ Position typically provided as cartesian coordinates ◮ Location is accurate to arbitrary precision ◮ Geographic routing protocols are designed and tested under
these conditions (mobile testbeds are hard)
◮ No basemap (from the routing protocol’s perspective) ◮ No WiFi MAC location database lookups
◮ We are as guilty of this as anyone
[Rohrer et al., 2008, Rohrer and Killeen, 2016]
www.nps.edu 11 / 39
Network Simulators
◮ Position typically provided as cartesian coordinates ◮ Location is accurate to arbitrary precision ◮ Geographic routing protocols are designed and tested under
these conditions (mobile testbeds are hard)
◮ No basemap (from the routing protocol’s perspective) ◮ No WiFi MAC location database lookups
◮ We are as guilty of this as anyone
[Rohrer et al., 2008, Rohrer and Killeen, 2016]
◮ What if we introduce error into simulation-reported locations?
www.nps.edu 11 / 39
Introduction Background Centroid-Based Routing Centroid Router Center-Mass Router Simulations Conclusion
www.nps.edu 12 / 39
Centroid Primitive
◮ Designed for simplicity ◮ Intended as building-block to be combined with other
techniques
Node Centroid
◮ Average position over time ◮ Updated at fixed time-interval
www.nps.edu 13 / 39
Introduction Background Centroid-Based Routing Centroid Router Center-Mass Router Simulations Conclusion
www.nps.edu 14 / 39
Centroid Routing Protocol
◮ Epidemic-based, multi-copy [Rohrer and Mauldin, 2018] ◮ Nodes exchange Centroid and message list at encounter ◮ Copy more messages to nodes with distant Centroid
www.nps.edu 15 / 39
Centroid Routing Protocol
◮ Epidemic-based, multi-copy [Rohrer and Mauldin, 2018] ◮ Nodes exchange Centroid and message list at encounter ◮ Copy more messages to nodes with distant Centroid ◮ Comparable complexity to Vector routing ◮ Improves efficiency by reducing message copying to nodes with
similar Centroids
◮ Intuition: Nodes with distant Centroids more likely to
encounter nodes not encountered by this node
www.nps.edu 15 / 39
Introduction Background Centroid-Based Routing Centroid Router Center-Mass Router Simulations Conclusion
www.nps.edu 16 / 39
Center-Mass Routing Protocol
◮ Builds on Centroid protocol ◮ Nodes record Centroid & timestamp of all encountered nodes ◮ Nodes exchange Centroid list and message list at encounter
◮ Merge into own Centroid list
◮ Copy more messages to nodes with distant Centroid ◮ Only copy messages to neighbors with Centroid closer to
destination Centroid
www.nps.edu 17 / 39
Center-Mass Routing Protocol
◮ Builds on Centroid protocol ◮ Nodes record Centroid & timestamp of all encountered nodes ◮ Nodes exchange Centroid list and message list at encounter
◮ Merge into own Centroid list
◮ Copy more messages to nodes with distant Centroid ◮ Only copy messages to neighbors with Centroid closer to
destination Centroid
◮ Intuition: Minimize message copies sent away from destination
www.nps.edu 17 / 39
Introduction Background Centroid-Based Routing Simulations Effects of Positional Sample Error Comparison of Probabilistic-Predictive Routing Protocols Protocol Efficacy Metric Conclusion
www.nps.edu 18 / 39
Network Simulator
The ONE [Keränen et al., 2009] is a discrete-time simulator designed specifically for DTN protocol simulations. It abstracts away details of propagation, medium access, and network layers and is written in Java.
◮ We use the default mobility and traffic scenario provided ◮ Commonly used Helsinki scenario in literature ◮ Nodes include pedestrians, cars, and trams ◮ Map-based mobility
www.nps.edu 19 / 39
Introduction Background Centroid-Based Routing Simulations Effects of Positional Sample Error Comparison of Probabilistic-Predictive Routing Protocols Protocol Efficacy Metric Conclusion
www.nps.edu 20 / 39
Simulated DTN Routing Protocols with GPS error
Vector [Kang and Kim, 2008]
◮ Flooding based ◮ Extrapolates trajectory from position history ◮ Limits message replication if nodes have similar trajectory
Centroid
◮ Flooding based ◮ Uses moving average of position history (centroid) ◮ Limits message replication if nodes have close centroids ◮ Original versions use raw node position from simulator ◮ We create a variant of each that adds random noise [±20 m]
to each position sample before using it (10% of typical noise)
www.nps.edu 21 / 39
Delivery Probability vs Bandwidth with GPS errors
102 103 104 transmit speed (kb/s) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 packet delivery ratio (PDR)
CenterMass CenterMassNoise Centroid CentroidNoise Vector VectorNoise
Figure: Effect of GPS errors on delivery probability vs radio bandwidth
www.nps.edu 22 / 39
Overhead vs Buffer Size with GPS errors
5 10 25 50 buffer size (MB) 20 40 60 80 100 120
Centroid CentroidNoise CenterMass CenterMassNoise Vector VectorNoise
Figure: Effect of GPS errors on overhead ratio vs buffer size
www.nps.edu 23 / 39
Overhead vs Bandwidth with GPS errors
102 103 104 transmit speed (kb/s) 20 40 60 80 100
VectorNoise Vector CentroidNoise Centroid CenterMassNoise CenterMass
Figure: Effect of GPS errors on overhead ratio vs radio bandwidth
www.nps.edu 24 / 39
Introduction Background Centroid-Based Routing Simulations Effects of Positional Sample Error Comparison of Probabilistic-Predictive Routing Protocols Protocol Efficacy Metric Conclusion
www.nps.edu 25 / 39
Comparison of Probabilistic-Predictive Routing Protocols
◮ Want to compare performance against DTN state-of-art ◮ Including non-geo protocols (Rapid, MaxProp, ProphetV2)
◮ (See paper for full list and citations)
◮ Also include oracle router for best-possible performance
benchmark
www.nps.edu 26 / 39
Delivery Probability vs Bandwidth 102 103 104 transmit speed (kb/s) 0.2 0.0 0.2 0.4 0.6 0.8 1.0 packet delivery ratio (PDR)
Oracle Gapr MaxProp Gapr2 Rapid CenterMass Centroid Vector ProphetV2
Figure: Delivery probability vs radio bandwidth broad comparison
www.nps.edu 27 / 39
Delivery Probability vs Buffer Size 5 10 15 20 25 30 35 40 45 50 buffer size (MB) 0.0 0.2 0.4 0.6 0.8 1.0 packet delivery ratio (PDR)
Oracle Gapr MaxProp Gapr2 Rapid Vector CenterMass Centroid ProphetV2
Figure: Delivery probability vs buffer size broad comparison
www.nps.edu 28 / 39
Overhead vs Buffer Size 5 10 25 50 buffer size (MB) 100 200 300 400 500 600
Centroid CenterMass Vector Gapr Gapr2 MaxProp Rapid ProphetV2
Figure: Overhead ratio vs buffer size broad comparison
www.nps.edu 29 / 39
Overhead vs Bandwidth 102 103 104 transmit speed (kb/s) 50 100 150 200 250 300
Rapid ProphetV2 MaxProp Gapr Vector Gapr2 Centroid CenterMass
Figure: Overhead ratio vs radio bandwidth broad comparison
www.nps.edu 30 / 39
Introduction Background Centroid-Based Routing Simulations Effects of Positional Sample Error Comparison of Probabilistic-Predictive Routing Protocols Protocol Efficacy Metric Conclusion
www.nps.edu 31 / 39
New efficacy metric for routing protocols
Efficacy = delivery ratio
(1)
◮ Captures trade-off between PDR/MDR and overhead ◮ Allows single-value comparison of protocols
www.nps.edu 32 / 39
Efficacy vs Buffer Size 5 10 25 50 buffer size (MB) 0.00 0.05 0.10 0.15 0.20 0.25 protocol efficacy
Centroid CenterMass Vector Gapr Gapr2 MaxProp Rapid ProphetV2
Figure: Protocol efficacy vs buffer size broad comparison
www.nps.edu 33 / 39
Efficacy vs Bandwidth 125 500 1000 5000 10000 transmit speed (kb/s) 0.00 0.05 0.10 0.15 0.20 0.25 protocol efficacy
Centroid CenterMass Vector Gapr Gapr2 MaxProp Rapid ProphetV2
Figure: Protocol efficacy vs radio bandwidth broad comparison
www.nps.edu 34 / 39
Introduction Background Centroid-Based Routing Simulations Conclusion
www.nps.edu 35 / 39
Takeaways
◮ Errors in GPS samples can significantly impact routing
performance
www.nps.edu 36 / 39
Takeaways
◮ Errors in GPS samples can significantly impact routing
performance
◮ Depending on how they are used ◮ Effects can be minimized through careful use of geolocation
data
◮ Such as Centroid/CenterMass routing primitives
◮ Efficacy helpful for protocol comparison
◮ Captures PDR/overhead tradeoff www.nps.edu 36 / 39
Thanks for watching!
www.nps.edu 37 / 39
Ben-Moshe, B., Elkin, E., Levi, H., and Weissman, A. (2011). Improving accuracy of GNSS devices in urban canyons. In Proceedings of the 23rd Canadian Conference on Computational Geometry (CCCG), pages 399–404. Kang, H. and Kim, D. (2008). Vector routing for delay tolerant networks. In Proceedings of the IEEE 68th Vehicular Technology Conference, pages 1–5. Keränen, A., Ott, J., and Kärkkäinen, T. (2009). The ONE simulator for DTN protocol evaluation. In Proceedings of the 2nd International Conference on Simulation Tools and Techniques (SIMUTools), pages 55:1–55:10, New York, NY, USA. ICST. Rohrer, J. P. (2018). Geographic centroid routing for vehicular networks. In Proceedings of the Seventh International Conference on Advances in Vehicular Systems, Technologies and Applications (VEHICULAR), Venice, Italy. IARIA. Rohrer, J. P., Jabbar, A., Perrins, E., and Sterbenz, J. P. G. (2008). Cross-layer architectural framework for highly-mobile multihop airborne telemetry networks. In Proceedings of the IEEE Military Communications Conference (MILCOM), pages 1–9, San Diego, CA, USA. Rohrer, J. P. and Killeen, K. M. (2016). Geolocation assisted routing protocols for vehicular networks. In Proceedings of the 5th IEEE International Conference on Connected Vehicles (ICCVE), pages 1–6, Seattle, WA. www.nps.edu 38 / 39
Rohrer, J. P. and Mauldin, A. N. (2018). Implementation of epidemic routing with ip convergence layer in ns-3. In Proceedings of the 2018 Workshop on ns-3 (WNS3), Surathkal, India. ACM. www.nps.edu 39 / 39