Upload
vodang
View
216
Download
2
Embed Size (px)
Citation preview
AutoMedia: Linking the Vehicle with Consumer Electronics and Services
Jason Flinn
University of Michigan
Problem: Vehicle is an islandEach vehicle is an island unto itself
data crosses the border generally with carried devicedevice (and its data) must be consciously managedlimited extra-vehicle networks are separate stovepipes
Several lost opportunitiespassengers don’t have what they wantcontext which could help organize is ignoredfew existing applications don’t coordinatemany vehicular applications hard to create
Jason Flinn
Vision: Bridge the islandsBridges to the “mainland” of home and office
unified view of all of a passenger’s mediaautomatic exchange of context and meta-dataallow individual consumer devices to come and gono effort on the part of the passengers
Bridges to the “other islands”—vehicles on the roadexchange traffic, road conditionscommunicate for immediate safety concernsform communities among co-located travelers
Jason Flinn
What’s needed?Advances in distributed storage:
Share data among home, vehicle, and web servicesCache data and propagate updates asynchronouslySupport non-traditional clients
Advances in networking:Leverage multiple networks: WiFi, WiMax, GPRS, DSRCPredict where you are goingProvide connectivity forecasts
Advances in middleware:Allow applications to declare intentionsHide vagaries of intermittent and uneven connectivity
Jason Flinn
What’s needed?Advances in distributed storage:
Shares data among home, vehicle, and mobile devicesCache data and propagate updates asynchronouslySupport non-traditional clients
Advances in networking:Leverage multiple networks: WiFi, WiMax, GPRS, DSRCPredict where you are goingProvide connectivity forecasts
Advances in middleware:Allow applications to declare intentionsHide vagaries of intermittent and uneven connectivity
Jason Flinn
Managing Data: Today
Jason Flinn
Managing Data: BlueFS
BlueFS helps manage personal data and devices
My Data
My Devices
Jason Flinn
BlueFS GoalsPervasive data access (anytime, anywhere, any device)
Support consumer electronics devices (iPod, phone, etc.)Support web services (Flickr, Facebook, Amazon S3, etc.)
Simplified data managementAutomation (indexing, organizing, transcoding, etc.)Policies (reliability, privacy, etc.)Context (location, time, nearby people, etc.)
Jason Flinn
BlueFS Overview
Server stores primary replica of all dataClients cache data to improve performanceClients register affinity for specific types of dataCached data available when disconnected
Jason Flinn
Device-Specific Protocol
CEDs cannot perform DFS protocol
DFS Protocol?
Connecting Consumer Electronics
Distributed File System
Device-Specific Protocol
Attach: computer speaks for CEDWorks with disconnected clients
DFS Protocol
Jason Flinn
Device-Specific NamespacesUser prefers one organization
Consumer electronics device requires another organization
Solution: Store translations on the consumer device
Classical
Jazz
F01
F02
F03
F04
Jason Flinn
Adding Automation
Updating indexes
Transcoding
Type-specific caching
Organizers
Jason Flinn
Adding Automation
Cannot execute on CED
Leverage general-purpose computersTake action when files change
Problem: Need notification of file changesJason Flinn
Notification Via Persistent Queries
Don’t need new mechanism
Leverage cache consistency mechanismStructure notifications as file system objectRobust to crashHandles disconnected operation
updates truncate
read process*.mp3
Jason Flinn
.TiVo
.mp4
Persistent QueriesExample: video recording and playback
FileServer
VideoTranscoder *.TiVo
.mp4
.mp4
.TiVo
*.TiVo .TiVo
.mp4
Jason Flinn
Leveraging intra-vehicle contextVehicle has a wealth of contextual data
Can we provide a “context bus” that enriches data?
Enrich data by sharing context among local devices:vehicle location used to tag site where photos taken
Allows searching by locationCan automatically organize digital photo albums
Wireless proximity of cell phones reveals user presenceMatch music on stereo to tastes of passengersStart playing audio-book where left off last time
Jason Flinn
What’s needed?Advances in distributed storage:
Shares data among home, vehicle, and mobile devicesCache data and propagate updates asynchronouslySupport non-traditional clients
Advances in networking:Leverage multiple networks: WiFi, WiMax, GPRS, DSRCPredict where you are goingProvide connectivity forecasts
Advances in middleware:Allow applications to declare intentionsHide vagaries of intermittent and uneven connectivity
Jason Flinn
The Big PictureState of the art: systems are reactive not proactive
There are opportunities to planpeople are creatures of habitinfrastructure is slow to change
If the system knewwhere you were likely to be, andwhat connectivity was available there...
...then applications could schedule their usage appropriately
Our goal: expose this connectivity derivative to applications
Jason Flinn
Pieces of the puzzleWhere are you likely to be?
figure out where you are nowuse the past to predict where you’ll be next
What are conditions likely to be when you get there?map access points at newly-visited locationsrefresh infrequently to catch stale state
Expose this information to applicationssimple API: expected bandwidth in 10 (20, 30, ...) secondsapplications use this however they like
Jason Flinn
Where are you?Existing solutions
GPS when you canWiFi localizationGSM fingerprinting
Our implementation: PlaceLab/WiFi20-30 meter avg accuracyacceptable for our needs
Convert to a discrete gridlimit precision of lat/longrid square ~110x80 meterscompact, useful representation
Jason Flinn
Where are you going?We need a personal mobility model
people are creatures of habit, but......different people have different habits
Song et. al. (Dartmouth) conducted a bake-offwinner: second-order Markov model
Model consists of discrete states and transitionsstates: your current and former locationtransitions: possible next states, with probabilities assignedprobabilities computed by past behaviordiscrete transition steps: we use 10 seconds (arbitrary)
Jason Flinn
What will conditions be like?
Overlapping public coverage
Heterogeneous qualityBandwidth, latency, ports
Decentralized ownership
Multiple “usable” options
Which AP is the best AP?
Jason Flinn
Virgil AP selection daemonCurrent mechanism: loudest must be best
okay in a homogenous deploymentno better than random selection in the wild
Instead, we perform active testing and discoveryScan for APs and associate to each in turnRun bandwidth, latency, port tests to reference serverChoose AP with the “best” connectionCache test results for future prediction
Jason Flinn
History is usefulRepeated traversals of the same area benefit (a lot!)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
Lap 1Lap 2Lap 3Lap 4Lap 5iwlist scan
Lap 1 Lap 2 Lap 3 Lap 4 Lap 50%
5%
10%
15%
20%
25%
randomSSSVirgil
History: % scans finding usable AP History: scanning overhead (s)
Jason Flinn
The Commute
Jason Flinn
Connectivity Forecasts
Close is often good enough, thanks to range issues
Jason Flinn
What’s needed?Advances in distributed storage:
Share data among home, vehicle, and mobile devicesCache data and propagate updates asynchronouslySupport non-traditional clients
Advances in networking:Leverage multiple networks: WiFi, WiMax, GPRS, DSRCPredict where you are goingProvide connectivity forecasts
Advances in middleware:Allow applications to declare intentionsHide vagaries of intermittent and uneven connectivity
Jason Flinn
Declarative NetworkingAllow applications to declare their intent in using network.
Main goal: simplicity of the interface.
Applications specify:Foreground (latency-sensitive) vs. backgroundBulk vs. small messagesImportant vs. unimportantetc.
May specify on a per-message, per-socket, or per-thread basis.
Jason Flinn
Connectivity middlewareManage vagaries in network connectivity:
Transient networks (WiFi hotspots)Intermittent networks (cell coverage)Changes in latencyVarying bandwidth
Challenge: map application sends/receives to networksChoose best network for each requestDefer traffic when necessary, continue when availableAllow applications to tear-down, re-establish connections
Jason Flinn
Declarative networking example
Read mail(foreground)
High availabilityLow bandwidth
IntermittentHigh bandwidth
Download mail(background)
Jason Flinn
Building BridgesAdvances in distributed storage:
Share data among home, vehicle, and mobile devicesCache data and propagate updates asynchronouslySupport non-traditional clients
Advances in networking:Leverage multiple networks: WiFi, WiMax, GPRS, DSRCPredict where you are goingProvide connectivity forecasts
Advances in middleware:Allow applications to declare intentionsHide vagaries of intermittent and uneven connectivity
Jason Flinn
Thanks to…
T.J. GiuliBrett HigginsAnthony NicholsonBrian NobleDan PeekVenkatesh PrasadAzarias RedaDavid Watson
Jason Flinn