Data Management Challenges and Opportunities in the Digital Home* ICME Amsterdam July 2005 Mike Franklin UC Berkeley *in collaboration with Intel Research Berkeley
Jan 20, 2016
Data Management Challenges and Opportunities in the Digital
Home*
ICMEAmsterdamJuly 2005
Mike FranklinUC Berkeley
*in collaboration with Intel Research Berkeley
Michael Franklin UC Berkeley EECS
Somewhere in Holland…
Michael Franklin UC Berkeley EECS
Data in the Home - Today
• Many sources and sinks
• Many media types, file formats
•“Outside” sources (e.g. CDDB, Tivo)
• Ad hoc, manual sharing/synching
• Minimal backup/archive support
• Manual organization, annotation, and search.
• Minimal sharing and integration across devices or applications.
Michael Franklin UC Berkeley EECS
Data in the Home - Where it’s Headed
• Standards enable new connections
• Even more sources and sinks
• Everything becomes “smart”
• Still no help with: backup, archive, organization, search, annotation, sharing, and integration.
• Who/What will manage all of this?
Michael Franklin UC Berkeley EECS
Is it a Networking Problem? – Audio
The devices depicted in these scenarios are for illustrative purposes onlyand have no relation to specific products planned by any manufacturer.
Server
From the Digital Home Working Group, 2004
Michael Franklin UC Berkeley EECS
Is it a Networking Problem? – Images
The devices depicted in these scenarios are for illustrative purposes onlyand have no relation to specific products planned by any manufacturer.
From the Digital Home Working Group, 2004
The devices depicted in these scenarios are for illustrative purposes onlyand have no relation to specific products planned by any manufacturer.
Michael Franklin UC Berkeley EECS
Is it a Networking Problem? – Video
The devices depicted in these scenarios are for illustrative purposes onlyand have no relation to specific products planned by any manufacturer.
From the Digital Home Working Group, 2004
The devices depicted in these scenarios are for illustrative purposes onlyand have no relation to specific products planned by any manufacturer.
Michael Franklin UC Berkeley EECS
Is it a Vendor-Specific Problem?
• PC and OS vendors - more powerful desktop machines with media-friendly OS’s.
• TV vendors • Set-top Box vendors• DVR vendors• Game Console vendors• Security System vendors• Home networking vendors• Home automation vendors
“Box Bias” - center of home is…
Michael Franklin UC Berkeley EECS
“A residence equipped with computing and information technology which anticipates and responds to the
needs of the occupants, working to promote their comfort, convenience, security and entertainment through
the management of technology within the home and connections to
the world beyond” Harper [2003]
“How smart does the bed in your house have to be before you are afraid to
go to sleep at night?” Rich Gold, The Plentitude
Is it an AI Problem?
Michael Franklin UC Berkeley EECS
• Multidisciplinary collaborations of Technologists, Ethnographers, Architects
• Sensors enable home to monitor:• Temperature• Light• Occupancy• Interactions?• Mood?
• Learning algorithms use measurements and feedback to predict occupant actions and needs.
Aware
Adaptive
Digital Home “Smart” Home?
Michael Franklin UC Berkeley EECS
The Aware Home
Michael Franklin UC Berkeley EECS
The Adaptive House
Michael Franklin UC Berkeley EECS
Current Status
• These and many other labs have helped push the research.• Although except for Moser’s Adaptive
House, they have not been really lived-in.
• But, smart home technology has been slow to make it to the mass market.
Michael Franklin UC Berkeley EECS
Our Approach
The home is becoming an increasingly data-intensive environment.
Point solutions will not scale.
A shared, data-centric infrastructure is needed.
A successful solution will enable “digital” home applications today, and provide a basis for “smart” home applications in the future.
Michael Franklin UC Berkeley EECS
What can we learn from Enterprise Data Management?
• Data Modeling - identifying and organizing entities and their relationships.
• Integration - combining disparate data.• Declarative Queries - set-based
languages for saying what you want, not how to get it.
• Indexing - accelerators for searching large data sets.
• Data Protection - Backup, Recovery, Archiving, Persistence, Consistency, Security.
Michael Franklin UC Berkeley EECS
So, it’s a Database Problem???
Michael Franklin UC Berkeley EECS
The Home is Different• No IT Staff to run it hands-off operation.
• Minimal IT budget must be cost-effective.
• User’s can and will reject it flexibility, adaptibility, context-awareness, “calmness”.
• People, families, homes, and contents change.
• Roles, needs, relationships not so clearly defined “SAP” for the home unlikely; privacy concerns are challenging.
Michael Franklin UC Berkeley EECS
Our Driving Applications• Preservation and location of digital information.
• Increasingly crucial data being stored on inherently short-lived devices. Want automatic backup, recovery, and caching.
• Tests: basic data management infrastructure, self-management.
• Energy management• Balance comfort and expense• Tests: sensor inputs, house temperature response models.
• Information displays - Home Portal• Example: InLook prototype
• Personalized news• Context-based media retrieval• State of family members, house, etc.
• Tests: Use of large/cheap displays, explore/demonstrate advantages of data integration.
Michael Franklin UC Berkeley EECS
Energy Management Application
Energy Management
Application Space
Infrastructure Space
Rules, Models, Archive
User/Environment Context
Calendar Information
$Pricing Signals
Michael Franklin UC Berkeley EECS
Home Portal - “InLook” (summer ‘04)
Dwell detector
Preferences
User context
Sensors
Michael Franklin UC Berkeley EECS
Hardware - The “data furnace”
Requirements:• Self- configuring, maintaining, tuning• Highly-reliable• Long life (~ 25 years)• Continually expandable/upgradable• Reasonable Cost
Goal: Invisible locus of control and reliable storage for the digital home. (not a PC)
No more cost or trouble than the home’s furnace.
Michael Franklin UC Berkeley EECS
Software Architecture
Discoverer(upnp)
appsLearning Engine
Bus
Mediagenerators
ArchiveQueries & Rules Sensors Actuators
• “Data-centric” view• Leverage our previous work on sensors and monitoring.
• Bus-based architecture for flexibility.• Central storage with caching at devices.
• Repository for Data and Metadata.• Repository for cross device/app Indexes.
Michael Franklin UC Berkeley EECS
UCB/IRB Digital Home Project
3 Challenges in Data Furnace Development
• Schema and Metadata
• Monitoring and Complex Event Processing
• Integrating Sensors
Michael Franklin UC Berkeley EECS
The Metadata Challenge
Need a model of:• People
•Family members and others.
•Roles, relationships,…
•Preferences• Home Layout• Devices & Data
Michael Franklin UC Berkeley EECS
Schema: Home, Place, Person, Event, Sensor
Some Issues:
• Model must evolve with the home and its members.
• Self-configuring: Cannot require significant human “start up” effort.
• Can such highly-personal entities such as homes be captured in a common schema?
Michael Franklin UC Berkeley EECS
Complex Event Processing
• Needed for monitoring and actuation.• Basis for system self-maintenance.• Key to prioritization (e.g., of detail data)• Can be implemented as simple
extensions to a streaming Query Language.
• Challenge: a single system that simultaneously handles events spanning seconds to years.
Michael Franklin UC Berkeley EECS
Data Stream Processing
QueriesQueries
Event SpecsEvent Specs
Subscrip-Subscrip-tionstions
QueriesQueries
Data
Traditional Database
Data Stream Processor
Result Tuples Result Tuples
•Data streams are unending
•Continuous, long-running queries
•Real-time processing
Data
http://telegraph.cs.berkeley.edu
Michael Franklin UC Berkeley EECS
Temporal Aggregation
SELECT S.room, AVG(temp)FROM SOME_STREAM S[range by ‘5 seconds’ slide by ‘5 seconds’]WHERE S.floor = ‘first’GROUP BY S.room
“I want to look at 5 seconds worth of data”
“I want a result tuple every 5 seconds”
A typical streaming query
Result Tuple(s)
Data Stream
Result Tuple(s)…
Window Clause
Michael Franklin UC Berkeley EECS
Spatial Aggregation
“I provide raw readings for an area”
“I provide avg values for a single room”
“I provide avg values for a floor”
“I provide avg values for the entire house”
• Continuous and Streaming
• Hierarchical• Coarser spatial and
temporal granularity as you go up?
• Some Issues• Automatic
placement and optimization
• Sharing of lower-level streams
Michael Franklin UC Berkeley EECS
Sensor-based Systems
• Receptors everywhere!• Wireless sensor networks, RFID technologies,
security systems, smart appliances, input devices ...
Need proper abstractions for dealing with varied devices
Michael Franklin UC Berkeley EECS
Metaphysical Data Independence
“Virtual Device(VICE)API”
Problem: how to deal with the complexity of physical devices?
http://hifi.cs.berkeley.edu
Michael Franklin UC Berkeley EECS
Integrating Heterogeneous Devices Using VICE: RFID & Sensor Motes
The Loudmouth Detector
Michael Franklin UC Berkeley EECS
The Virtues of VICE
• Once you have the right abstractions:• Soft Sensors (e.g., a “person detector”)• Quality and lineage streams• Pushdown of external validation information• Power management and other
optimizations• Data Archiving• Model-based sensing• “Non-declarative” code• …
Michael Franklin UC Berkeley EECS
Putting it all Together
• We are proposing a data-centric view towards digital home infrastructure.
• The goal is to adapt enterprise-class data management techniques to the home.• Non-trivial differences between home and
enterprise.
• Currently focused on: • Data modeling for the home.• Self-managing hardware and software platforms
using complex event processing and continuous queries.
• Sensor integration using the VICE API.• We are also strengthening our
collaborations with ethnographers and architects.
Michael Franklin UC Berkeley EECS
Conclusions
via Anind Dey (CMU)
Our message: Home is where the bits are…
Michael Franklin UC Berkeley EECS
Acknowledgements
This is joint work with the Digital Home project at UC Berkeley and Intel Research Berkeley, and the UC Berkeley Database Group:
•Ryan Aipperspach
•Kurt Brown
•John Canny
•Lilia Gutnik
•Wei Hong
•Allison Woodruff
•Gustavo Alonso
•Shawn Jeffery
•Sailesh Krishnamurthy
•Shariq Rizvi