Querying the Physical World: The COUGAR Project

1
Querying the Physical WorldThe COUGAR Project

• Philippe BonnetJohannes Gehrke
• Praveen Seshadri
• CORNELL University

2
Outline

• 1 Device networks represent a new and exciting
  application domain for database technology
• 2 The COUGAR device database system transfers
  existing techniques in this new context
• 3 Device Database Systems raise new challenges

3
Device-based applications

• Environmental applications
• Flood detection, volcano monitoring, plant
  breeding,
• Surveillance applications
• Tactical information systems, building
  surveillance, Centrale Nationale dAlarme,
• Control applications
• Traffic control in a large city, warehouse
  monitoring,

4
Catalog of Devices
5
Are Devices Clients or Servers?

• Scenario 1
• DB server hold primary copy of the data
• Devices replicate data
• Devices support user interaction
• Scenario 2
• Devices hold primary copy of the data
• Users access large collection of devices

6
Classification of Devices
7
Wireless Integrated Network Sensor
Sensoria.com WINS NG Architecture
8
Wireless Sensor Network Characteristics

• Cost of communication is orders of magnitude
  higher than cost of local computation on devices
• Cost of communication with neighbor devices is
  lower than cost of communication with distant
  devices

9
Access to Device Networks

• Access to large collection of devices
• Associative access independent of the physical
  organization
• Users formulate queries regardless of the device
  network organization

10
Declarative Access to Device Networks

• Aggregate Queries over Historical Data
• Give me the average rainfall level in California
  for 1999
• Snapshot queries about the current state of the
  device network
• What is the current rainfall level in California?
• Long-running queries for monitoring the device
  network
• Tell me whenever two rainfall sensors in the same
  area detect an abnormal rainfall level.

11
Database Approaches for accessing Device Networks

• Warehousing approach
• Device data is extracted in a predefined way
• Device data is stored in a centralized DB server
• Queries are evaluated on the centralized DB
  server
• Distributed approach
• Queries are evaluated by contacting devices
• Portions of queries are executed on the devices

12
Device Database System

• Access to devices is dictated by the query
  workload
• Trade-off between local computation on devices
  and reduced communication

13
Candidate Techniques (1)

• (Re-)active database systems
• Active rules to implement propagation
• Rich expressive power of active rules is not
  required
• No mechanisms for execution of active rules in a
  distributed context
• Continuous queries over append-only relations
• Mechanism for propagation of insertions
  integrated with SQL query execution
• Defined in a centralized context

14
Candidate techniques (2)

• Distributed Database Systems
• Techniques for balancing local computation and
  communication
• Assume that a limited number of full-fledged
  database systems is tightly interconnected
• No support for long-running queries
• Mediator Systems
• Represent devices as data sources with possibly
  limited capabilities
• Limited range of join execution techniques in
  existing systems, data from distinct sources is
  always joined on the mediator

15
The COUGAR Approach

• How to represent devices?
• Schema
• Internally
• How to formulate long-running queries?
• How to evaluate long-running queries in a
  distributed context?

16
Devices as Abstract Data Types

• A device provides a set of functions
• Synchronous continuous phenomenon
• Asynchronous threshold events
• To each device type is associated an Abstract
  Data Type
• Device ADT method represent device functions
• e.g., getTemperature() detectTempGreaterThan(90)

17
Device ADT Methods as Virtual Relations

• R.dev.getTemperature(90) R
  VRGetTemperature

• Base relation
• Located on front-end
• Virtual relation
• Append-only
• Partitioned across devices

18
Long-running queries

• Snapshot and long-running queries expressed in
  SQL with slight modification of the language
• WHERE clause event condition
• SELECT clause action
• New statements for creation, deletion and
  evaluation (for a given duration) of long-running
  queries
• User-defined function for representing interval
  between successive invocation of device functions
• Implicit conditions on timestamped relations

19
Examples of Long-running queries

• CREATE LR_QUERY q1 AS
• SELECT R.dev, R.dev.getTemperature()
• FROM TempSensors R, NamedPlaces N
• WHERE every(30)
• AND R.dev.location().inside(N.bbox) AND
  N.name California
• CREATE LR_QUERY q2 AS
• SELECT R1.dev.location()
• FROM TempSensors R1, TempSensors R2
• WHERE every(0)AND R1.dev.detectAbnormalTemperatu
  re()AND R2.dev.detectAbnormalTemperature()AND
  R1.dev R2.dev

20
Query Execution Plan

• Virtual Scan Blocking read operator to implement
  propagation of insertions in (append-only)
  fragments of virtual relations
• Similar to fetch_wait operator defined in Alert
• Distributed joins involving
• A base relation and a virtual relation
• Two virtual relations

21
Query Execution Plan

• Constraints on joins involving virtual scans
• Join on partitioned relations
• Virtual relations cannot be scanned repeatedly
• Index join (or dependent join) is required if
  virtual relation input attribute is bound
• No materialization of virtual relations

22
Distributed Join Techniques

• Base Relation Virtual Relation
• With or without semi-join reduction

BR
VRn
VRn
BR
VRi
VRi
BR
VR2
VR2
VR1
VR1
BR
BR
23
Cost Model

• Metrics
• Reaction Time
• Resource Usage
• Cost in joules Wcpu CPU Wram RAM Wsend
  NbMsgs Wbdw SizeMsgs
• Wcpu 0.000001 J/instruction
• Wram 0
• Wsend 0.059 J/msg
• Wbdw 0.23 J/ Kbytes

24
Some new Challenges

• Decentralized query execution/optimization
• A single site cannot maintain precise
  meta-information about the complete system
• Adaptative query processing
• Conditions in a device network change over time
• Devices fail, move or disconnect

25
The Cornell COUGAR Project

• A first version of the Cougar system is
  implemented on top of the WINS sensor network
  infrastructure.
• Demonstration at the Intel Computing Continuum
  Conference
• http//www.intel.com/intel/cccon
• Demonstration in the context of the DARPA Sensit
  Program
• http//www.darpa.mil/ito/research/sensit/