1. Query Processing over a Sensor Network Cornell University Johannes Gehrke Philippe Bonnet

2. Database Server Frontend Overview Sensor.com WINS NG Architecture

3. Problems Data Model: How to represent devices in a uniform way so that queries can be asked to access them? Architecture: How to connect the database server and the devices? Query Execution: Can the traditional query execution techniques be applied in a device database system? Query Optimization: Are traditional cost based optimizer well suited for device database systems?

4. Problems Data Model: How to represent devices in a uniform way so that queries can be asked to access them? Architecture: How to connect the database server and the devices? Query Execution: Can the traditional query execution techniques be applied in a device database system? Query Optimization: Are traditional cost based optimizer well suited for device database systems?

5. Problems Data Model: How to represent devices in a uniform way so that queries can be asked to access them? Architecture: How to connect the database server and the devices? Query Execution: Can the traditional query execution techniques be applied in a device database system? Query Optimization: Are traditional cost based optimizer well suited for device database systems?

6. Problems Data Model: How to represent devices in a uniform way so that queries can be asked to access them? Architecture: How to connect the database server and the devices? Query Execution: Can the traditional query execution techniques be applied in a device database system? Query Optimization: Are traditional cost based optimizer well suited for device database systems?

7. Data Model Abstract Data Type (ADT) : class of devices and their methods ex: Infrared Detector : IR IR.detectTrackedVehicle() IR.detectPersonel() IR.signature() IR.latitude() IR.longitude() IR.powerStatus()IR.TurnDown()IR.TurnOn() ex: Camera Camera.orientateTowards(int X, int Y) Camera.latitude()Camera.longitude() Relation : collection of objects of the same class

8. Monitoring Queries SELECT R.IR.latitude(), R.IR.longitude() FROM InfraRedDetectors R WHERE R.IR.detectTrackedVehicle(); SELECT R.IR FROM InfraRedDetectors R, Positions P WHERE $near(R.IR.latitude(), R.IR.longitude(), P.latitude, P.longitude) AND P.name = “Marina Del Rey Hotel” AND R.IR.detectPersonel(); Declarative Queries

9. Control Queries SELECT R.IR.turnDown() FROM InfraRedDetectors R WHERE R.IR.latitude() < 100; SELECT C.Camera.orientateTowards(X,Y) FROM Camera C, InfraredDetectors R WHERE $near(R.IR.latitude(), R.IR.longitude(), C.Camera.latitude(), C.Camera.longitude()) AND R.IR.detectTrackedVehicle();

10. Architecture Predator Proxy Server Proxy SQL query Graphical User Interface Sensor Network Proxy on each node Proxy Server needs to address groups of proxies: at least by class of device possibly more expressive predicates Proxy Server Proxy

11. Query Execution Step 1 –Query Processing on the database server (Database Server currently implemented Proxies on Windows CE devices - 01/2000) –Performance Evaluation Step 2 –Distribution of Query Fragments to the proxy servers and proxies (01/2001)

12. Query Optimization Relies on meta-information describing the sensor network: –categories of devices, methods supported on each device, statistics concerning the usage of devices, distribution criterion,... Initial Optimizer: 01/2000 Server-based Optimizer: 09/2000 Distributed Optimizer:09/2001

13. Conclusion Today –Prototype based on Predator data model, architecture (proxy servers and proxies in Java), query execution, draft query optimization. Period 2: –Proxies on Win CE+WINS NG –Server-based Optimizer.