In-Network Query Processing on Heterogeneous Hardware Martin Lukac*†, Harkirat Singh*, Mark Yarvis*, Nithya Ramanathan*† *Intel Research †CENS TRIANGLE! SELECT SHAPE(covered) FROM sensors : Triangle! 11010: Triangle! PCMCIA Card Slot Serial Port Power Connector & Switch JTAG Interface USB Port Ethernet Port Input Button Reset Switch Mote Interface CF Card Slot Integrates into sensor networks through mote interface Expandability through PCMCIA, Serial, USB, and CF Card slots Sensors (web cams, GPS) Communication (802.11, Ethernet) Storage (Flash memory) In-Network Processing on Heterogeneous Hardware Targeting class of applications which require more processing or storage capability These application typically observe the environment for signatures and require FFT’s or database lookups Shape detection is representative of storage heavy applications: database of sensor values –> shapes Shape Detection Grid of Mica2 motes detects shape of object placed on top Additions to TinyDB covered attribute uses light sensor to determine if mote is covered SHAPE aggregate merges sensor vales on a mote SHAPE aggregate calls EmStar shape component on a Stargate Heterogeneous Routing is used to designate the Stargate as a more powerful node EmStar shape component looks up sensor values in database and returns list of likely matches Shape result merged to aggregate data and forwarded Mica2 Mote (TinyOS) TinyOS Transciever WSN EmStar Components and Services Stargate (Linux) TinyOS Application EmToS MoteNIC (Emstar Serial Link) EmStar S Attraction gradient Packet Forwarder Routing Metric Service Value MUX Route Table Heterogeneous Forwarding TinyDB Query Routing Query Procesing Service Value MAC/PHY Std. Interface Control Energy Reliability Latency Adjusts routing metric Update Handler Heterogeneous RoutingTinyDB 1.TinyDB – Queries network for desired signature 2.Heterogeneous Routing – Attracts data to Stargate 3.Stargate – Provides processing and storage for query 4.EmStar & EmTOS – Runs same code as motes and provides calls to the Linux Environment TinyDB + Heterogeneous Routing Packet routing should reflect network resources E.g. Packet aggregation, filtering Transparently adjust routes to attract data to resource-rich nodes Balance the value of in-network processing against the cost of packet forwarding Proactive attraction and aggregation of data reduces overall forwarding cost Implemented on top of DSDV Stargate Micro-Server The Stargate platform developed at Intel Research 400MHz Intel XScale ® Processor (PXA255) 64 MB SDRAM, 32 MB FLASH Full Linux software environment Easily integrate TinyOS Apps with other languages, libraries, and tools EmTOS EmTOS is the bridge between TinyOS and EmStar Emulates TinyOS API Runs unmodified NesC code TinyOS apps can interact with EmStar components and the entire Linux environment Distributed debugging and development environment for Linux Targeted to sensor network app development Offers different degrees of “reality”, from simulation to deployment Same code in all realities EmStar Enable query processing on a subset of nodes Attract packets to query processing nodes TinyDB specifies the “value” of query processing to the network layer Value depends on data reduction “effectiveness” of query Database abstraction for sensor networks Query network with SQL variant Queries are temporal User develops applications as queries instead of NesC and TinyOS code TinyDB provides the data management functions Query optimization, power efficient execution Improve efficiency through the use of aggregates: MAX, SUM, AVG, … Tree based routing for query delivery, data, collection, and in-network aggregation D B S C A E F SELECT nodeid, temp FROM sensors WHERE temp > 40 EPOCH DURATION 6s Result route Known links Initial Query 58D3 47C3 48E2 42F2 57D2 46C2 47E1 55D1 45C1 TempNodeEpoch Matlab GIS * Other names and brands may be claimed as the property of others.