Vehicular cloud computing: from intelligent transport to urban surveillance

Vehicular cloud computing: from intelligent transport to urban surveillance VCA 2012 Cyprus, June 19, 2012 Mario Gerla UCLA, Computer Science Dept

Mobile Cloud Computing (MCC) Workshop! Colocated with Sigcomm 2012, Helsinki! Workshop date: August 17, 2012! Organizers:! Mario Gerla (University of California Los Angeles)" Dijiang Huang (Arizona State University)"

Outline! From Internet Clouds to Mobile Clouds! Types of mobile clouds" Principles and Challenges" Mobile Computing platform design tools" The evolution of vehicular communications! Initial applications: safety and content distrinution" Two Vehicle Cloud Applications:! Urban surveillance" Vehicular Traffic Management" Tools for Mobile Computing Cloud support! UCLA Vehicle Campus Testbed! Future Directions!!

The Mobile Cloud! Internet Cloud (eg Amazon, Google etc)! Immense computer, storage resources" BroadBand Connectivity" Services, security" Mobile Cloud Computing (traditional)! What most researchers mean:" Access to the Internet Cloud from mobiles" Tradeoffs between local and cloud computing (eg m-health)" Recently, new Mobile Cloud paradigm:! Mobile nodes increasingly powerful (storage, process, sensors)" Emerging distributed applications not suited for Amazon" Enter: Mobile Computing Cloud (MCC)!

Several types of Mobile Clouds! Vehicular Cloud! Intelligent Transport applications" Surveillance" Opportunistic formation" Personal Cloud! Info and advertisment dissemination" Mobile health apps" Dependent on the Internet cloud" Influenced by social networks" Mission Oriented Cloud! Not opportunistic ; rather planned for a goal" Scouting of a particular terrain"

Observed trends: Vehicular Cloud 1. Vehicles perform increasingly more complex (sensor) data collection/processing road alarms (pedestrian crossing, electr. brake lights, etc) cooperative content downloading via P2P car- torrent surveillance (video, mechanical, chemical sensors) road mapping via crowd sourcing accident, crime witnessing (for forensic investigations, etc) 2. Spectrum is scarce => Internet upload expensive With Vehicular Cloud Computing: Keep and process data on vehicle cloud instead of uploading to Internet cloud

Example of Vehicular Cloud Vehicles in the same geographic domain form a P2P cloud and engage in collaborative activities P2P communications leveraging spectrum holes in the urban unlicensed spectrum Inter-cloud communications via Infrastructure (3G, WiFI) food and gas info. Related work: MobiCloud Dijian Huang Maui MSR Auton Vehi Clouds-S. Olariu IC Net On Wheels Fan Bai GM regulating entrance to the evacuation highway

Vehicle Cloud vs Internet Cloud Both offer a significant pool of resources:! computing, storage, communications" However:! Vehicular applications are mainly location relevant! Data Sources: Drivers or environment" Customers: generally the drivers " Vehicle cloud interacts with:! Internet cloud" Personal (smart phone) cloud" Exception: emergency VANET! During an emergency the infrastructure is destroyed"! Emergency cloud is isolated, works on its own power" Vehicles (especially E-vehicles) become the ONLY available computing/ comms resource"

Vehicle Cloud computing - classification Cloud computing classification: Sensing the environment Communications to exchange data Processing of on-board data Interaction with Internet Cloud (and edge cloud)

Vehicle Cloud Support Tools! Content Based Addressing! CCX, NDN, Haggle etc" Efficient Dissemination/propagation! Synchronization! Collaborative Team formation! Spectrum sensing; cog. frequency selection! Security/Privacy; authentication; revocation!

Roadmap ahead Evolution of Vehicular Communications Emerging Vehicle Cloud applications Surveillance Monitoring the environment events (natural or manmade) Urban Traffic management: From totally centralized to locally distributed to evacuation

The Vehicle Transport Challenge! Safety! 33,963 deaths/year (2009)" 5,800,000 crashes/year " Leading cause of death for ages 4 to 34! Mobility! 4.2 billion hours of travel delay" $78 billion cost of urban congestion" Environment! 2.9 billion gallons of wasted fuel" 22% CO 2 from vehicles"

In 2003 DOT launches: Vehicle Infrastr. Integration (VII)! VII Consortium: USDOT, automakers, suppliers,..! Goal: V2V and V2I comms protocols to prevent accidents! Technology validation;" Business Model Evaluation" Legal structure, policies" Testbeds: Michigan, Oakland (California)! However: 10 year to deploy 300,000 RSUs and install DSRC on 100% cars! Meanwhile: can do lots with 3G and smart phones! Can we speed up proof of concept?! Enter Connected Vehicle (2009-2014)!

The Connected Vehicle Program! Previous Veh Infrastr Integr (VII) Model (2003-9)! DSRC based for all applications" Start with V2I (for all application types) and evolve into V2V (safety) " Current DOT s Connected Vehicle Plan! Non-safety (mobility, environment)! Leverage existing data sources & communications; include DSRC as it becomes available" Safety DSRC! Aggressively pursue V2V; " Can leveraging of nomadic devices & retrofitting accelerate benefits?"

Standard: DSRC / IEEE 802.11p/ WAVE! Dedicated 5.9Ghz spectrum! Derived from 802.11a!! Three types of channels: V2V service, V2I service and control broadcast channel! IEEE 1609.x: channel allocation, channel estimation, power, data rate control, etc! Forward radar Display Event data recorder (EDR) Positioning system Computing platform Communication facility Rear radar

Emerging Application Areas! Safe Navigation! Location Relevant Content Distr.! Urban Sensing! Efficient, Intelligent, Clean Transport!

! " V2V for Safe navigation! Forward Collision Warning,! Intersection Collision Warning.! Advisories to other vehicles about road perils! Ice on bridge, Congestion ahead,."!

Car to Car communications for Safe Driving! Vehicle type: Cadillac XLR Curb weight: 3,547 lbs Speed: 75 mph Acceleration: + 20m/sec^2 Coefficient of friction:.65 Driver Attention: Yes Etc. Alert Status: None Vehicle type: Cadillac XLR Curb weight: 3,547 lbs Speed: 65 mph Acceleration: - 5m/sec^2 Coefficient of friction:.65 Driver Attention: Yes Etc. Alert Status: None Alert Status: Inattentive Driver on Right Alert Status: Slowing vehicle ahead Alert Status: Passing vehicle on left Vehicle type: Cadillac XLR Curb weight: 3,547 lbs Speed: 75 mph Acceleration: + 10m/sec^2 Coefficient of friction:.65 Driver Attention: Yes Etc. Alert Status: Passing Vehicle on left Vehicle type: Cadillac XLR Curb weight: 3,547 lbs Speed: 45 mph Acceleration: - 20m/sec^2 Coefficient of friction:.65 Driver Attention: No Etc.

Location relevant content delivery! Traffic information! Local attractions, advertisements! Tourist information, etc!! CarTorrent : cooperative download of location multimedia files! " "

You are driving to Vegas You hear of this new show on the radio Video preview on the web (10MB)

One option: Highway Infostation download Internet file

Incentive for opportunistic ad hoc networking Problems: Stopping at gas station for full download is a nuisance Downloading from GPRS/3G too slow and quite expensive 3G broadcast services (MBMS, MediaFLO) only for TV Observation: many other drivers are interested in download sharing Solution: Co-operative P2P Downloading via Car-Torrent (like Bit Torrent in the Internet)!!!!!

CarTorrent: Basic Idea! Internet Download a piece Outside Range of Gateway Transferring Piece of File from Gateway

Co-operative Download: Car Torrent! Internet Vehicle-Vehicle Communication Exchanging Pieces of File Later

Simulation Results! Completion time density! 200 nodes 40% popularity Time (seconds)

Surveillance Scenario: storage and retrieval! Participating Cars ( busess, taxicabs, commuters):! Continuously collect images from the street (store data locally)" Process the data and detect an event" Classify the event as Meta-data (Type, Option, Loc, time,vehicle ID)" Question: how to access this info?! - Sensing - Processing Summary Harvesting CRASH Crash Summary Reporting Meta-data : Img, -. (10,10), V10

Mobility-assisted Meta-data Diffusion/ Harvesting (Mobeyes 2007)! HREP HREQ Agent harvests a set of missing meta-data from neighbors Periodical meta-data broadcasting + Broadcasting meta-data to neighbors + Listen/store received meta-data

MobEyes: Mobility-assisted Epidemic Dissemination! Mobeyes exploit mobility to disseminate metadata!! Source periodically broadcasts meta-data to neighbors! Only source can advertise meta-data" Neighbors store advertisements in their local memory" Drop stale data" A mobile agent (the police) harvests meta-data from vehicles by querying them (with Bloom filter)!

Urban Surveillance as a Service! Suppose a green truck was involved in a suicidal terrorist attack! The truck blows up " The Agency wants to find out the approach path of the vehicle" Wants to learn if any accomplices drove along with the truck up to the target" Conventional solution: road side video cameras! Sophisticated attacker can avoid video cameras, or disable them" Mobeyes based solution:! Reconstruct the path by using the other vehicles video cameras" We have run a simulation experiment in a 2kmx2km area, with 100 vehicles roaming! Vehicles watch each other and agent reconstructs all paths"

Trace Reconstruction!

Trace Reconstruction!

Intelligent navigation!! GPS Based Navigators! Dash Express (came to market in 2008):! Synergy between Navigator Server and City Transport Authority!

NAVOPT Navigator Assisted Route Optimization!! On Board Navigator! Interacts with the Server" Periodically transmits GPS and route" Receives route instructions" Manhattan grid (10x10)! 5 routes (F1~ F5) from source to destination" Link capacity: 14,925 [vehicles/h]" F2,5 F3,4 S F5 F2 Shortest path F1 F3 D F4

Analytic Results! Total average delay (h/veh)! 0.45" 0.4" 0.35" Average delay (hour)! 0.3" 0.25" 0.2" 0.15" shortest path" flow deviation" 0.1" 0.05" 0" 13500" 13600" 13700" 13800" 13900" 14000" 14100" 14200" 14300" 14400" 14500" 14600" 14700" 14800"

Analytic Results (cont)! Average speed (Km/h)! 70" 60" 50" Speed (Km/h)! 40" 30" shortest path" flow deviation" 20" 10" 0" 13500" 13600" 13700" 13800" 13900" 14000" 14100" 14200" 14300" 14400" 14500" 14600" 14700" 14800"

Sumo simulation results! Sumo-0.12! 10 X10 grid" Road segment: 400m" Length of vehicle: 4m" Max speed limit: 60Km/h" Average delay! Delay increases drastically around 15000 rate [veh/h] in case of shortest path" In NAVOPT, delay slightly increases around 20000"

What happens when the grid gets congested?! As density increases, flow decreases! Eventually the system comes to a stand still (bottleneck)" The gradient flow optimization (eg Flow Deviation) works only in free flow mode (below congestion density level)! Function no longer convex => many local minima" In congested mode, new techniques must be used" Question: how can the Navigator Server detect that the system is congested and how badly?! Must correlate and interpret the space-time probes obtained by vehicles"

Lessons Learned! When density increases beyond critical value, the mode changes from FREE FLOW to CONGESTED! Phase transition delayed by V2V communications" The Navigator Server can detect congestion because the velocity on the road segment is uneven:! Random, distributed slow down => Shockwaves" Localized traffic jam => backlog propagation" If the system is congested, basic gradient method optimization does not work:! Function is not convex => multiple local optima" Need to exercise FLOW CONTROL! "

Nav Opt Next Steps! Make the process {traffic measmt and route enforcement} iterativly! Deal with Free Flow as well as Backlogged! SUMO + Qualnet will be used as the simulation platform! Include pollution component based on engine regime! Include pollution costs in the optimization!

Distributed traffic management! Centralized traffic management cannot react promptly to local traffic perturbations! A doubled parked truck in the next block; A traffic accident; A sudden surge of traffic" Internet based Navigator Server cannot micro-manage traffic for scalability reasons " Enter distributed, v-cloud based traffic mgmt! Distributed approach a good complement of centralized supervision" On the Effectiveness of an Opportunistic Traffic Management System for Vehicular Networks, Leontiadis et al, IEEE Trans on ITS Dec 2011!!

CATE: Comp Assisted Travl Environment! Vehicles crowd source traffic information and traffic load data base:! 1) sensing traffic information from road segment travel time;" 2) sharing it with neighboring vehicles (in an ad hoc manner); and" 3) dynamically recomputing the best route to destination from the current position." The study was done by simulation:! QUALNET a popular event driven MANET simulator, and" MobiDense, a mobility simulator that combines topology and traffic flow information to generate a mobility trace." Case Study: Traffic pattern for Portland obtained from Los Alamos Lab" Potential limitations of CATE:! Delay in traffic loads propagation; lack of trip destination info"

Traffic loading w/o CATE! Green no congest Yellow moderate Red heavy congest"

Traffic loading with CATE! Green no congest Yellow moderate Red heavy congest"

Histogram of vehicle decreased trv time!

Information Propagation Speed! Two-dimensional Heatmap of age of received information (in seconds) about the link highlighted by the arrow (bridge)."

CATE tested on C-VET! Up to 8 vehicles roaming the Campus with GPS, WiFi radios and 250m range! Static throughput between two nodes = 30Mps! At 30km/h througput = 7Mbps! Propagation of a 2MB block (traffic sample) from one node to the other 7 nodes:! First vehicle received full block in 20s" Next four in < 72s" Last two in < 125s " C-VeT experiments are comparable with Portland simulation results!

Integrating Centralized and Distributed traffic management! Central Navigator Server (in the Amazon cloud):! Provides MACRO traffic instructions" Is aware of user destinations" Accounts for possible congestion fees" Can perform ECO-Routing (accounting for pollution)" Interacts with City Traffic/Planning Department (traffic lights, Green waves, access ramp control)" Distributed (CATE-like) traffic management in the Vehicular Cloud:! Can handle sudden traffic jams, accidents, other anomalies" Provides miopic traffic redirections w/o preempting Server" Can be a safety net when infrastructure fails" Amzon Cloud + Vehicle Cloud :! Improves scalability, reaction time, robustness to disasters"

Vehicle Cloud Support Tools! Content Based Addressing! CCX, NDN, Haggle etc" Efficient Dissemination! Synchronization! Collaborative Team formation! Spectrum sensing; cog. frequency selection! Security/Privacy!

CBMEN ENCODERS DARPA CBMEN Program Content Based MANET - Edge Networking Mark-Oliver Stehr, (SRI) David Anhalt, Hua Li (SET) José Joaquin Garcia-Luna-Aceves (UCSC)! Mario Gerla (UCLA) 49

Finding content in a MANET! MANET is ideally suited for emergency & tactical operations, but!! hard to find content" no search engine" Goal: Scalable, efficient content search in MANETs Content Based Networking! 50

Network Coding to assist content retrieval! File/1 Interest triggers download of content File/1 Content is encoded in three blocks File/2 File/3 File/2 File/3 File/1 File/2 File/3 File (coded) File (coded) File (coded) Get File/1, File/2, File/3 at once 51

Applying Cognitive Radios to Wi-Fi based Vehicular Networks

Real Time Surveillance in Urban Motorcade! Urban motorcade:! Surveillance video multicast!escort cars transmit video streams to agents in motorcade using WiFi! Problem: urban WiFi is becoming increasingly congested! Approach: Use Cognitive Radios to minimize impact on urban WiFI! presidential motorcade

Westwood experiment (4 blocks) IEEE 802.11b/g! 50 Mhz 227APs discovered using wireless card in automatic detection mode. Orthogonal channels 1,6,11 are predominant

Urban WiFi Spectrum Sharing Problem! Secondary users! Escorts generate surveillance video streams and multicast to patrollers" Store and forward within motorcade only" Primary network residential WiFi users! Dense WiFi access point deployment must minimize disruption of residential WiFi" In turn, minimize interference caused by residential traffic on motorcade"

On-Demand Multicast Routing Protocol (ODMRP)! S F F R F F On-demand mesh creation! A source initiates Join Query flooding when it has data to send." Intermediate nodes relay Join Query after recording the previous hop as backward pointer." Multicast subscribers send Join Reply messages, following backward pointers to the source." Upon receiving a Join Reply, a node declares itself as part of the forwarding group." R R R R mesh Join Query Join Reply

CoCast: Cognitive Multicast! Join Query flooding in the Common Control Channel! Cognitive nodes compute Active channels (two-way available) after receiving Join Queries that contain Available channels.! Available! 3, 6, 7! Available! 1, 5, 7! Active! Listen! Transmit! S" Available! 1, 2, 4, 5! Active! Listen! Transmit! 1,5,7 7 data channels (1~7), 1 common control channel (0) Available! 1, 2, 5, 6! Active! Listen! Transmit! 1, 5 F" 1, 2, 5 F" Available! 1, 2, 4, 6! Active! 1, 2, 6 Listen! Transmit! 1,2,5,6 F" F" Available! 2, 4, 6! Active! 2, 6 Listen! Transmit! 1,2,4,6 R" R" R" Active! 6 Listen! Transmit! Available! 1, 2, 6, 7! Active! 1, 2, 6 Listen! Transmit! Available! 1, 2, 6! Active! 2, 6 Listen! Transmit!

CoCast: Channel Allocation! Sending back Join Reply! Receiver selects Listen channel from Active channel list, and reports in JR" Forward nodes set own TX Channel to Listen channels of downstream node" Goal: minimize # of channel frequency switches along path" Available! 3, 6, 7! Available! 1, 5, 7! Active! Listen! Transmit! 1 S" Available! 1, 2, 4, 5! Active! 1, 5! Listen! Transmit! 1 2 1 F" 2 Available! 1, 2, 5, 6! Active! 1, 2, 5! Listen! Transmit! 2 2, 6 F" 6 2 Available! 1, 2, 4, 6! Active! 1, 2, 6! Listen! Transmit! F" F" Available! 2, 4, 6! Active! 2, 6! Listen! Transmit! 2 2 6 6 6 2 6 R" R" R" Active! 6! Listen! 6 Transmit! Available! 1, 2, 6, 7! Active! 1, 2, 6! Listen! 6 Transmit! Available! 1, 2, 6! Active! 2, 6! Listen! 2 Transmit!

Simulation Setup! Simulation configuration in QualNet 3.9! Parameter" Value" Number of Nodes! 100! Mobility! Switching delay (D switch )! Number of channels! Channel data rate! Packet size/rate! RGM model, 20m/s! 0~50 ms! 7 Data channels, 1 CCC! 2 Mbps! 512 Bytes, 4pps! Reference Group Mobility Model! Random way point motion within the motorcade" Each geographic area has different WiFi channel occupancies! Channel idle/busy state changes in space and time"

Simulation Results! Variable number of video sources! 2 ~ 10 source nodes within the same motorcade" Comparison between ODMRP and CoCast in terms of delivery ratio! Static case vs. mobile case "" CoCast scales better with increasing number of source nodes! Nice surprise: frequency switching along the path mitigates HIDDEN TERMINAL"

SORA SDR Kit High-performance Software Defined Radio kit from MSR Asia. Exploits multiple cores available in CPU. 3+ stations available. Plans Implementation of Cognitive Protocols OFDM channel measurements OFDM subcarrier assignment 61

C-Ve T Campus - Vehicular Testbed E. Giordano, A. Ghosh,! G. Marfia, S. Ho, J.S. Park, PhD! System Design: Giovanni Pau, PhD! Advisor: Mario Gerla, PhD!!

The Deployment! We plan to install our node equipment in:! 30 Campus operated vehicles (including shuttles and facility management trucks). " Exploit on a schedule and random campus fleet mobility patterns " 30 Commuting Vans: Measure urban pollution, traffic congestion etc" 12 Private Vehicles: controlled motion experiments " Cross campus connectivity:10 node WiFi Mesh + 2 WiMAX base stations" Support: NSF GENI, ARMY DURIP, CISCO!

C-VeT Goals! Provide:!! A shared virtualized environment to test new protocols! Full Virtualization! MadWiFi Virtualization (with on demand exclusive use)" Multiple OS support (Linux, Windows)" Assisted by:! Qualnet Simulator " VEMU fully virtualized Emulator" Enable:! Propagation experiments! Collection of mobility traces and network statistics! Use of platform for Urban Sensing, Geo routing etc! Deployment of innovative V2V/V2I applications!

Preliminary Experiments! Equipment:! Up to 6 cars roaming the UCLA Campus" 802.11abg radios" Routing protocol: OLSR" " Experiments:! Connectivity map computed by OLSR" Azureus P2P application" Propagation model validation" OLSR and AODV comparison" Crowd Sourcing to discover urban grid details (eg position of traffic lights, their cycle, etc)" "

Line of sight and Corner propagation models! Q. Sun et al, Analytical formulae for path loss prediction in urban street grid.., IEEE TVT, July 2005.

CORNER: fixed to mobile!

C-VeT Experiments in Progress!! Cognitive radios to improve spectrum utilization! Virtualized Router experiments! Multiple routing protocols (OLSR, AODV)" Multiple OS support (Linux, Windows). " Soft Handoff experiments (WiFI, WiMAX)! Multipath with Network Coding" Propagation model validation for 5.9 Ghz (DSRC frequency band)! Intelligent Navigation experiments! Dissemination delays"

Work ahead in Vehicle Cloud research Vehicle cloud formation and maintenance Subclouds with common social interests In-Cloud networking Cloud Content dissemination, storage, indexing, search (Content Centric Networking) Efficient spectrum usage (via cog radios) Exploit WiFi spectrum short range Exploit TV spectrum long range connections (to other vehicle clouds) User-centric security and privacy protection Interaction with Internet Cloud Interaction with pedestrian (people) clouds

The End! Thank You!