New trends in pervasive computing: vehicular sensor platforms ISWPC 2006 Phuket, Jan Mario Gerla Computer Science Dept UCLA

Outline Car to Car communications Car Torrent Ad Torrent Network games Cars as mobile sensor platforms

Pervasive Computing: conventional view Ubiquitous, “unconscious” user/sensor interaction Sensors/actuators are embedded in the environment The user (fixed or mobile) interacts with the sensors - reads them, sets them and computes –Sensors on the airport walls drive traveler to the proper gate –Wall sensors/actuators turn on lights, air conditioning, appliances.. –Firefighters interact with preinstalled sensors/RFIDs to plan next step during disaster recovery

New “pervasive” paradigm: sensors in motion Sensors on mobile platforms (eg, cars, people, robots, UAVs) The user must extracts the information from a moving set of “observers” –Car sensor platforms: a posteriori “event” reconstruction in the vehicular grid - eg, car accident, kidnapping, terrorist attack, etc –Instrumented soldiers, firefighters etc: gaining situation awareness at the command post

Pervasive Computing and Mobility Mobility helps: –Cars can see a lot more “situations” because they are moving –Also, as they move, they provide uniform coverage and opportunistic deployment –They can propagate, “disseminate” the info in the environment more efficiently –Moving sensors are more difficult to temper with and disable However, mobility also creates new challenges: –Difficult to reconstruct space and time correlation –If I am looking for data collected in a specific street, at a specific time, where do I find it? –Also connectivity to mobiles may break Other Car to Car applications exhibit similar behavior

The themes of this talk Introduce and compare various car to car scenarios Identify the building blocks for efficient car to car application design investigate tradeoffs between opportunities and challenges introduced by mobility

Outline Opportunistic ad hoc networking –Cars interact with each other and with environment via ad hoc wireless networking –Ad hoc networking among cars is “opportunistic”, much less predictable than in traditional large scale systems such as the battlefield Car to car Content sharing/delivery –Car torrent –Ad torrent The car sensor application –Disaster recovery (eg, chemical spill, terrorist attack) –Accident reconstruction –Traffic condition reporting

Opportunistic ad hoc networking

What is an opportunistic ad hoc net? wireless ad hoc extension of the wired/wireless infrastructure coexists with/bypasses the infrastructure generally low cost and small scale Examples –Indoor W-LAN extended coverage –Group of friends networked with Bluetooth to share an expensive resource (eg, 3G connection) –Peer to peer networking in the urban vehicle grid

Traditional ad hoc nets –Civilian emergency, tactical applications –Typically, large scale –Instant deployment –Infrastructure absent (so, must recreate it) –Very specialized mission/function (eg, UAV scouting behind enemy lines) –Critical: scalability, survivability, QoS, jam protection –Non critical: Cost, Standards, Privacy

Opportunistic ad hoc nets –Commercial, “commodity” applications –Mostly, small scale –Cost is a major issue (eg, ad hoc vs 2.5 G) –Connection to Internet often available –Need not recreate “infrastructure”, rather “bypass it” whenever it is convenient –Proximity routing instead of long range routing –Critical: Standards are critical to cut costs and to assure interoperability –Critical: Privacy, security is critical

Car to Car ad networking is “opportunistic” Most Cars applications can in principle connect to the internet (via WiFi, 3G, satellite etc) and download data entirely from it In fact direct Internet connection (if available) should always be considered as an option However, cost, delay and wireless channel load are the limiting factors

Why opportunistic ad hoc networking? All Internet access will soon be wireless Most Internet terminals will be mobile with multiple radio interfaces (WiFi, Bluetooth, 3G etc) Yet, “single hop” access from terminal may not be feasible, or may not be efficient! –Obstacles; distance –Cost –Inefficient use of resources –Proximity networking application –etc Enter Opportunistic multi-hop networking

Urban “opportunistic” ad hoc networking From Wireless to Wired network Via Multihop

Opportunistic piggy rides in the urban mesh Pedestrian transmits a large file block by block to passing cars, busses The carriers deliver the blocks to the hot spot

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

DSRC*/IEEE p : Enabler of Novel Applications Car-Car communications at 5.9Ghz Derived from a three types of channels: Vehicle-Vehicle service, a Vehicle-Gateway service and a control broadcast channel. Ad hoc mode; and infrastructure mode p: IEEE Task Group that intends to standardize DSRC for Car-Car communications * DSRC: Dedicated Short Range Communications

DSRC Channel Characteristics

Hot Spot Vehicular Grid as Opportunistic Ad Hoc Net

Hot Spot Power Blackout Power Blackout Vehicular Grid as Emergency Net

Power Blackout Power Blackout Vehicular Grid as Emergency Net

Car to Car Data Sharing Applications

CarTorrent : Opportunistic Ad Hoc networking to download large multimedia files Alok Nandan, Shirshanka Das Giovanni Pau, Mario Gerla WONS 2005

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 Observation: many other drivers are interested in download sharing (like in the Internet) Solution: Co-operative P2P Downloading via Car-Torrent

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

Co-operative Download Vehicle-Vehicle Communication Internet Exchanging Pieces of File Later

Bit Torrent review Swarming: Parallel downloads among a mesh of cooperating peers –Scalable: System capacity increases with increase in number of peers Tracker –Handles peer discovery Centralized Tracker, single point of failure Observation: Might not work for Wireless scenarios, because of intermittent connectivity Issue: Mobility increases churn of nodes participating in a download

BitTorrent…. A picture.. Uploader/downloader Tracker Uploader/downloader

Experimental Evaluation

CarTorrent: Gossip protocol A Gossip message containing Torrent ID, Chunk list and Timestamp is “propagated” by each peer Problem: how to select the peer for downloading

Peer Selection Strategies Possible selections: 1) Rarest First: BitTorrent-like policy of searching for the rarest bitfield in your peerlist and downloading it 2) Closest Rarest: download closest missing piece (break ties on rarity) 3) Rarer vs Closer: weighs the rare pieces based on the distance to the closest peer who has that piece.

Impact of Selection Strategy

Why is the Car-Torrent solution attractive? Bandwidth at the infostation is limited and “not convenient” –It can become congested if all vehicles stop –It is a nuisance as I must stop and waste time GPRS and 3G bandwidth is also limited and expensive The car to car bandwidth on the freeway is huge and practically unlimited! Car to car radios already paid for by safe navigation requirement CarTorrent transmissions are reliable - they involve only few hops (proximity routing) Question: does mobility help or hurt?

AdTorrent: Digital BillBoards for Vehicular Networks V2V COM Workshop Mobiquitous 2005 WONS 2006 Alok Nandan, Shirshanka Das Biao Zhou, Giovanni Pau, Mario Gerla

Digital Billboard Safer : Physical billboards can be distracting to drivers Less intrusive : The skyline is not marred by unsightly boards. Efficient : With a “filter” on the client (vehicle) side, users will see the Ad only if they actively search for it or are interested in it. Localized : The physical wireless medium automatically induces locality characteristics into the advertisements.

Digital Billboard Every Access Point (AP) disseminates Ads that are relevant to the proximity of the AP From simple text-based Ads to trailers of nearby movies, virtual tours of hotels etc Business owners in the vicinity subscribe to this digital billboard service for a fee. Need a location-aware distributed application to search, rank and deliver content to the end-user (the vehicle)

AdTorrent Features Keyword Set Indexing to reduce Communication Overhead Epidemic Scoped Query Data Dissemination – optimized for vehicular ad hoc setting Torrent Ranking Algorithm Swarming in actual content delivery Discourage Selfishness

AdTorrent : Search and Content Delivery How is it different from CarTorrent? –Push model for data dissemination as opposed to pull only (by queries) as in CarTorrent –Search for relevant content in the vehicular network Keyword Set Indexing for communication efficient search Search query dissemination optimized for vehicular networks Models for evaluating impact of search query hit w.r.t. Scope of epidemic query dissemination

Keyword Set Indexing KeyURL of DataMetaData Hash(AB) Size = 10MB Hash(BC) Genre= 0xAB Hash(AC) Hash(CD) A, B, C and D are individual keywords Performs Remote “AND’ing” of index entries Why is it communication efficient?

Query Dissemination and Lookup Search local cache for Ads/Torrents Scoped flood –Scoping level is a key design parameter –Tradeoff between reliability and communication overhead Bloom Filter of cached index entries exchanged Finally, when the query dissemination and response is done, the querying node has a list of index entries that match.

Torrent Ranking Algorithm Which torrent/ads to pick if multiple query hits? –Proximity –Maximum number of pieces –“Stability” of neighbors –Relevance of keywords to MetaData queried

Hit Rate vs. Hop Count with LRU

Summary We presented two P2P applications for Vehicular Wireless Networks Peer Discovery and Torrent Ranking uses proximity Content Delivery –Push and Pull models –Swarming Communication efficient search and gossip protocols System Designers can use our analytical framework to estimate the required cache size and scope of query dissemination based on the user performance requirements Hop limit of 4 gives an adequate hit rate and the incremental gain of increasing the scope of query dissemination was minimal in the mobility scenarios used.

Vehicular Sensor Network (VSN) PerSens 2006 Uichin Lee, Eugenio Magistretti (UCLA) Infostation Car-Car multi-hop 1. Fixed Infrastructure 2. Processing and storage 1. On-board “black box” 2. Processing and storage Car to Infostation

Vehicular Sensor Network (VSN) - Applications Systems engineering –Road surface diagnosis/maintenance –Traffic pattern analysis Environment –Urban environmental pollution monitoring Homeland security –Imaging from streets or recording sounds for accident or crime site investigation (terrorist alerts) Commercial applications –“Mobile” bazaar service (large-scale p2p)

VSN Scenario: storage and retrieval Private Cars: –Continuously collect images on the street (store data locally) –Process the data and detect an event –Classify the event as Meta-data (Type, Option, Location, Vehicle ID) –Post it on distributed index Police retrieve data from distributed storage Crash! Meta-data : Img, -. (10,10), V10 Meta-data : Img, Crash, (10,5), V12

Searching on Mobile Storage - Building an Index Major tasks: Post / Harvest Naïve approach: “Flooding” –Not scalable to thousands of nodes (network collapse) –Network can be partitioned (data loss) Design considerations –Non-intrusive: must not disrupt other critical services such as inter- vehicle alerts –Scalable: must be scalable to thousands of nodes –Disruption or delay tolerant: even with network partition, must be able to post & harvest “meta-data” Proposed approach: Distributed index

Distributed Index options Info station based index “Epidemic diffusion” index –Mobile nodes periodically broadcast meta-data of events to their neighbors (via epidemic diffusion) –A mobile agent (the police) queries nodes and harvests events –Data may be dropped when temporally stale and geographically irrelevant

Epidemic Diffusion - Idea: Mobility-Assist Data Diffusion

1) “periodically” Relay (Broadcast) its Event to Neighbors 2) Listen and store other’s relayed events into one’s storage Keep “relaying” its meta-data to neighbors

Epidemic Diffusion - Idea: Mobility-Assist Data Harvesting Meta-Data Req 1.Agent (Police) harvests Meta-Data from its neighbors 2.Nodes return all the meta-data they have collected so far Meta-Data Rep

VSN: Mobility-Assist Data Harvesting (cont) Assumption –N disseminating nodes; each node n i advertises event e i “k”-hop relaying (relay an event to “k”-hop neighbors) –v: average speed, R: communication range –ρ : network density of disseminating nodes –Discrete time analysis (time step Δ t) Metrics –Average event “percolation” delay –Average delay until all relevant data is harvested

VSN: Simulation Simulation Setup –Implemented using NS-2 –802.11a: 11Mbps, 250m transmission range –Average speed: 10 m/s –Network: 2400m x 2400m –Mobility Models Random waypoint (RWP) Road-track model (RT) : Group mobility model with merge and split at intersections –Westwood map is used for a realistic simulation

Road Track Mobility Model

Simulation Simulation Setup –Implemented using NS-2 –802.11a: 11Mbps, 250m tx range –Average speed: 10 m/s –Network: 2400m*2400m –Mobility Models Random waypoint (RWP) Real-track model: –Group mobility model –merge and split at intersections (RT) Westwood map is used for simulations

Event diffusion delay: Random Way Point 1. ‘k’-hop relaying 2. m event sources Fraction of Infected Nodes K=1,m=1 K=1,m=10 K=2,m=10 K=2,m=1

Event diffusion delay: Road Track 1. ‘k’-hop relaying 2. m event sources Fraction of Infected Nodes

Data harvesting delay with RWP Agent Regular Nodes

Harvesting results with Road Tracks AGENT REGULAR

Conclusions Introduced several car to car applications: –Dynamic content sharing/delivery: Car Torrent (Pull model) Ad Torrent (Push+ Pull model) –Pervasive, mobile sensing (push model) New Research Challenges: – “Opportunistic” ad hoc network paradigm: Epidemic dissemination Proximity routing –Searching massive mobile storage –Realistic mobility models (waypoint mobility not enough!) –Delay tolerant networking –Security, privacy –Reward Third Party forwarding; prevent “cheating” Future Goal: define common framework for C2C applications

Related Projects UMassDiesel (UMass) –A Bus-based Disruption Tolerant Network (DTN) – VEDAS (UMBC) –A Mobile and Distributed Data Stream Mining System for Real-Time Vehicle Monitoring and diagnostics – CarTel (MIT) –Vehicular Sensor Network for traffic conditions and car performance – RecognizingCars (UCSD) –License Plate, Make, and Model Recognition –Video based car surveillance –

The End Thank You