tinydb tutorial

1

Implementation and Research Issues in Query Processing for Wireless

Sensor Networks

Wei Hong Intel Research, Berkeley

[email protected]

Sam MaddenMIT

[email protected]

ICDE 2004

2

Motivation• Sensor networks (aka sensor webs, emnets) are here

– Several widely deployed HW/SW platforms• Low power radio, small processor, RAM/Flash

– Variety of (novel) applications: scientific, industrial, commercial

– Great platform for mobile + ubicomp experimentation

• Real, hard research problems to be solved– Networking, systems, languages, databases

• We will summarize:– The state of the art– Our experiences building TinyDB– Current and future research directions

Berkeley Mote

3

Sensor Network Apps

Traditional monitoring apparatus.

Earthquake monitoring in shake-test sites.

Vehicle detection: sensors along a road, collect data about passing vehicles.

Habitat Monitoring: Storm petrels on Great Duck Island, microclimates on James Reserve.

4

Declarative Queries

• Programming Apps is Hard– Limited power budget– Lossy, low bandwidth communication– Require long-lived, zero admin deployments– Distributed Algorithms– Limited tools, debugging interfaces

• Queries abstract away much of the complexity– Burden on the database developers– Users get:

• Safe, optimizable programs• Freedom to think about apps instead of details

5

TinyDB: Prototype declarativequery processor

• Platform: Berkeley Motes + TinyOS• Continuous variant of SQL : TinySQL

• Power and data-acquisition based in-network optimization framework

• Extensible interface for aggregates, new types of sensors

6

Agenda

• Part 1 : Sensor Networks (50 Minutes)– TinyOS– NesC

• Short Break• Part 2: TinyDB (1 Hour)

– Data Model and Query Language– Software Architecture

• Long Break + Hands On• Part 3: Sensor Network Database

Research Directions (1 Hour, 10 Minutes)

7

Part 1

• Sensornet Background• Motes + Mote Hardware

– TinyOS– Programming Model + NesC

• TinyOS Architecture– Major Software Subsystems– Networking Services

8

A Brief History of Sensornets

• People have used sensors for a long time• Recent CS History:

– (1998) Pottie + Kaiser: Radio based networks of sensors

– (1998) Pister et al: Smart Dust• Initial focus on optical communication• By 1999, radio based networks, COTS Dust, “Motes”

– (1999) Estrin + Govindan• Ad-hoc networks of sensors

– (2000) Culler/Hill et al: TinyOS + Motes– (2002) Hill / Dust: SPEC, mm^3 scale computing

• UCLA / USC / Berkeley Continue to Lead Research•Many other players now•TinyOS/Motes as most common platform

• Emerging commercial space: • Crossbow, Ember, Dust, Sensicast, Moteiv, Intel

9

Why Now?

• Commoditization of radio hardware– Cellular and cordless phones, wireless

communication

• Low cost -> many/tiny -> new applications!

• Real application for ad-hoc network research from the late 90’s

• Coming together of EE + CS communities

10

Motes

4Mhz, 8 bit Atmel RISC uProc

40 kbit Radio

4 K RAM, 128 K Program Flash, 512 K Data Flash

AA battery pack

Based on TinyOS

Mica MoteMica Mote

Mica2DotMica2Dot

11

History of Motes

• Initial research goal wasn’t hardware– Has since become more of a priority with

emerging hardware needs, e.g.:• Power consumption• (Ultrasonic) ranging + localization

– MIT Cricket, NEST Project• Connectivity with diverse sensors

– UCLA sensor board

– Even so, now on the 5th generation of devices• Costs down to ~$50/node (Moteiv, Dust)• Greatly improved radio quality• Multitude of interfaces: USB, Ethernet, CF, etc.• Variety of form factors, packages

12

Motes vs. Traditional Computing

• Lossy, Adhoc Radio Communication

• Sensing Hardware• Severe Power Constraints

13

Radio Communication

• Low Bandwidth Shared Radio Channel– ~40kBits on motes– Much less in practice

• Encoding, Contention for Media Access (MAC)

• Very lossy: 30% base loss rate– Argues against TCP-like end-to-end

retransmission• And for link-layer retries

• Generally, not well behaved

From Ganesan, et al. “Complex Behavior at Scale.” UCLA/CSD-TR 02-0013

14

Types of Sensors

• Sensors attach via daughtercard

•Weather–Temperature–Light x 2 (high intensity PAR, low intensity, full spectrum)–Air Pressure–Humidity

•Vibration–2 or 3 axis accelerometers

•Tracking–Microphone (for ranging and acoustic signatures)–Magnetometer

• GPS

15

Power Consumption and Lifetime

• Power typically supplied by a small battery– 1000-2000 mAH– 1 mAH = 1 milliamp current for 1 hour

• Typically at optimum voltage, current drain rates

– Power = Watts (W) = Amps (A) * Volts (V)– Energy = Joules (J) = W * time

• Lifetime, power consumption varies by application– Processor: 5mA active, 1 mA idle, 5 uA sleeping– Radio: 5 mA listen, 10 mA xmit/receive, ~20mS / packet– Sensors: 1 uA -> 100’s mA, 1 uS -> 1 S / sample

16

• Each mote collects 1 sample of (light,humidity) data every 10 seconds, forwards it

• Each mote can “hear” 10 other motes• Process:

– Wake up, collect samples (~ 1 second)– Listen to radio for messages to forward (~1

second)– Forward data

Power Consumption Breakdown

0

10

20

30

40

50

60

70

80

90

Radio Sensors Processor

Hardware Element

Percentage of Total Power

Energy Usage in A Typical Data Collection Scenario

Processor Energy Breakdown

05

101520253035404550

Idle Waiting

for Radio

Waiting

for

Sensors

Sending

Processing Phase

Percentage of Total Energy

17

Sensors: Slow, Power Hungry, NoisyTime of Day vs. Light

-20

0

20

40

60

80

100

120

140

160

180

200

20:09 20:38 21:07 21:36 22:04 22:33 23:02 23:31 0:00 0:28 0:57 1:26

Time of Day

Lux

Chamber Sensor

Sensor 69

Time of Day vs. Light

-20

0

20

40

60

80

100

120

140

160

180

200

20:09 20:38 21:07 21:36 22:04 22:33 23:02 23:31 0:00 0:28 0:57 1:26

Time

Light (Lux)

Chamber Sensor

Sensor 69 (Median of Last 10)

18

Programming Sensornets: TinyOS

• Component Based Programming Model

• Suite of software components– Timers, clocks, clock synchronization– Single and multi-hop networking– Power management– Non-volatile storage management

19

Programming Philosophy

• Component Based– “Wiring” to components together via

interfaces, configurations

• Split-Phased– Nothing blocks, ever.– Instead, completion events are signaled.

• Highly Concurrent– Single thread of “tasks”, posted and

scheduled FIFO– Events “fired” asynchronously in response

to interrupts.

20

NesC

• C-like programming language with component model support– Compiles into GCC-compatible C

• 3 types of files:– Interfaces

• Set of function prototypes; no implementations or variables– Modules

• Provide (implement) zero or more interfaces• Require zero or more interfaces• May define module variables, scoped to functions in module

– Configurations• Wire (connect) modules according to requires/provides

relationship

21

Component Example: Leds

module LedsC { provides interface Leds;}implementation{ uint8_t ledsOn;

enum { RED_BIT = 1, GREEN_BIT = 2, YELLOW_BIT = 4 };

…. async command result_t Leds.redOn() { dbg(DBG_LED, "LEDS: Red on.\n"); atomic { TOSH_CLR_RED_LED_PIN(); ledsOn |= RED_BIT; } return SUCCESS; }….}

22

Configuration Example

configuration CntToLedsAndRfm {}implementation { components Main, Counter, IntToLeds, IntToRfm, TimerC;

Main.StdControl -> Counter.StdControl; Main.StdControl -> IntToLeds.StdControl; Main.StdControl -> IntToRfm.StdControl; Main.StdControl -> TimerC.StdControl; Counter.Timer -> TimerC.Timer[unique("Timer")]; IntToLeds <- Counter.IntOutput; Counter.IntOutput -> IntToRfm;}

23

Split Phase Examplemodule IntToRfmM { … }implementation { …command result_t IntOutput.output (uint16_t value) { IntMsg *message = (IntMsg *)data.data; if (!pending) { pending = TRUE; message->val = value; atomic { message->src = TOS_LOCAL_ADDRESS; } if (call Send.send(TOS_BCAST_ADDR, sizeof(IntMsg), &data)) return SUCCESS; pending = FALSE; } return FAIL; }

event result_t Send.sendDone (TOS_MsgPtr msg,

result_t success) { if (pending && msg == &data) { pending = FALSE; signal IntOutput.outputComplete (success); } return SUCCESS; }}

}

24

Major Components

• Timers: Clock, TimerC, LogicalTime

• Networking: Send, GenericComm, AMStandard, lib/Route

• Power Management: HPLPowerManagement

• Storage Management: EEPROM, MatchBox

25

Timers

• Clock: Basic abstraction over hardware timers; periodic events, single frequency.

• LogicalTime: Fire an event some number of H:M:S:ms in the future.

• TimerC: Multiplex multiple periodic timers on top of LogicalTime.

26

Radio Stack• Interfaces:

– Send• Broadcast, or to a specific ID• split phase

– Receive• asynchronous signal

• Implementations:– AMStandard

• Application specific messages• Id-based dispatch

– GenericComm• AMStandard + Serial IO

– Lib/Route• Mulithop

IntMsg *message = (IntMsg *)data.data;…message->val = value;atomic { message->src = TOS_LOCAL_ADDRESS;}call Send.send(TOS_BCAST_ADDR, sizeof(IntMsg), &data))

event TOS_MsgPtr ReceiveIntMsg.receive(TOS_MsgPtr m) {

IntMsg *message = (IntMsg *)m->data; call IntOutput.output(message->val); return m; }

Wiring to equate IntMsg to ReceiveIntMsg

27

Multihop Networking

• Standard implementation “tree based routing”

A

B C

D

FE

B B

B

BB

B

B

B

B

B BB

R:{…}

R:{…}

R:{…}

R:{…} R:{…}

Problems:

Parent SelectionAsymmetric LinksAdaptation vs. Stability

Node DNeigh QualB .75C .66E .45F .82

Node CNeigh QualA .5B .44D .53F .35

28

Geographic Routing• Any-to-any routing via geographic

coordinates– See “GPSR”, MOBICOM 2000, Karp + Kung.

A

B

•Requires coordinate system*

•Requires endpont coordinates

•Hard to route around local minima (“holes”)

*Could be virtual, as in Rao et al “Geographic Routing Without Coordinate Information.” MOBICOM 2003

29

Power Management

• HPLPowerManagement– TinyOS sleeps processor when possible– Observes the radio, sensor, and timer state

• Application managed, for the most part– App. must turn off subsystems when not in use– Helper utility: ServiceScheduler

• Peridically calls the “start” and “stop” methods of an app

– More on power management in TinyDB later– Approach works because:

• single application• no interactivity requirements

30

Non-Volatile Storage

• EEPROM– 512K off chip, 32K on chip– Writes at disk speeds, reads at RAM speeds– Interface : random access, read/write 256 byte

pages– Maximum throughput ~10Kbytes / second

• MatchBox Filing System– Provides a Unix-like file I/O interface– Single, flat directory– Only one file being read/written at a time

31

TinyOS: Getting Started

• The TinyOS home page:– http://webs.cs.berkeley.edu/tinyos– Start with the tutorials!

• The CVS repository– http://sf.net/projects/tinyos

• The NesC Project Page– http://sf.net/projects/nescc

• Crossbow motes (hardware):– http://www.xbow.com

• Intel Imote– www.intel.com/research/exploratory/motes.htm.

32

Part 2

The Design and Implementation of TinyDB

33

Part 2 Outline

• TinyDB Overview• Data Model and Query Language• TinyDB Java API and Scripting• Demo with TinyDB GUI• TinyDB Internals• Extending TinyDB• TinyDB Status and Roadmap

34

TinyDB RevisitedSELECT MAX(mag) FROM sensors WHERE mag > threshSAMPLE PERIOD 64ms

• High level abstraction:– Data centric programming– Interact with sensor

network as a whole– Extensible framework

• Under the hood:– Intelligent query

processing: query optimization, power efficient execution

– Fault Mitigation: automatically introduce redundancy, avoid problem areas

App

Sensor Network

TinyDB

Query, Trigger

Data

35

Feature Overview

• Declarative SQL-like query interface• Metadata catalog management• Multiple concurrent queries• Network monitoring (via queries)• In-network, distributed query processing• Extensible framework for attributes,

commands and aggregates• In-network, persistent storage

36

TinyDB GUI

TinyDB Client APIDBMS

Sensor network

Architecture

TinyDB query processor

0

4

0

1

5

2

6

3

7

JDBC

Mote side

PC side

8

37

Data Model

• Entire sensor network as one single, infinitely-long logical table: sensors

• Columns consist of all the attributes defined in the network

• Typical attributes:– Sensor readings– Meta-data: node id, location, etc.– Internal states: routing tree parent, timestamp, queue

length, etc.• Nodes return NULL for unknown attributes• On server, all attributes are defined in catalog.xml• Discussion: other alternative data models?

38

Query Language (TinySQL)

SELECT <aggregates>, <attributes>

[FROM {sensors | <buffer>}][WHERE <predicates>][GROUP BY <exprs>][SAMPLE PERIOD <const> |

ONCE][INTO <buffer>][TRIGGER ACTION <command>]

39

Comparison with SQL

• Single table in FROM clause• Only conjunctive comparison predicates

in WHERE and HAVING• No subqueries• No column alias in SELECT clause• Arithmetic expressions limited to

column op constant• Only fundamental difference: SAMPLE

PERIOD clause

40

TinySQL Examples

SELECT nodeid, nestNo, lightFROM sensorsWHERE light > 400EPOCH DURATION 1s

1EpocEpoc

hhNodeiNodei

ddnestNnestN

ooLightLight

0 1 17 455

0 2 25 389

1 1 17 422

1 2 25 405

Sensors

“Find the sensors in bright nests.”

41

TinySQL Examples (cont.)

Epoch region CNT(…) AVG(…)

0 North 3 360

0 South 3 520

1 North 3 370

1 South 3 520

“Count the number occupied nests in each loud region of the island.”

SELECT region, CNT(occupied) AVG(sound)

FROM sensors

GROUP BY region

HAVING AVG(sound) > 200

EPOCH DURATION 10s

3

Regions w/ AVG(sound) > 200

SELECT AVG(sound)

FROM sensors

EPOCH DURATION 10s

2

42

Event-based Queries

• ON event SELECT …• Run query only when interesting events

happens• Event examples

– Button pushed– Message arrival– Bird enters nest

• Analogous to triggers but events are user-defined

43

Query over Stored Data

• Named buffers in Flash memory• Store query results in buffers• Query over named buffers• Analogous to materialized views• Example:

– CREATE BUFFER name SIZE x (field1 type1, field2 type2, …)

– SELECT a1, a2 FROM sensors SAMPLE PERIOD d INTO name

– SELECT field1, field2, … FROM name SAMPLE PERIOD d

44

Using the Java API

• SensorQueryer– translateQuery() converts TinySQL string into

TinyDBQuery object– Static query optimization

• TinyDBNetwork– sendQuery() injects query into network– abortQuery() stops a running query– addResultListener() adds a ResultListener that is

invoked for every QueryResult received– removeResultListener()

• QueryResult– A complete result tuple, or– A partial aggregate result, call mergeQueryResult()

to combine partial results• Key difference from JDBC: push vs. pull

45

Writing Scripts with TinyDB

• TinyDB’s text interface– java net.tinyos.tinydb.TinyDBMain –

run “select …”– Query results printed out to the

console– All motes get reset each time new

query is posed• Handy for writing scripts with shell,

perl, etc.

46

Using the GUI Tools

• Demo time

47

Inside TinyDB

TinyOS

Schema

Query Processor

Multihop Network

Filterlight >

400get (‘temp’)

Aggavg(tem

p)

QueriesSELECT AVG(temp) WHERE light > 400

ResultsT:1, AVG: 225T:2, AVG: 250

Tables Samples got(‘temp’)

Name: tempTime to sample: 50 uSCost to sample: 90 uJCalibration Table: 3Units: Deg. FError: ± 5 Deg FGet f : getTempFunc()…

getTempFunc(…)getTempFunc(…)

TinyDBTinyDB

~10,000 Lines Embedded C Code

~5,000 Lines (PC-Side) Java

~3200 Bytes RAM (w/ 768 byte heap)

~58 kB compiled code

(3x larger than 2nd largest TinyOS Program)

48

Tree-based Routing

• Tree-based routing– Used in:

• Query delivery • Data collection• In-network aggregation

– Relationship to indexing?

A

B C

D

FE

Q:SELECT …

Q Q

Q

QQ

Q

Q

Q

Q

Q QQ

R:{…}

R:{…}

R:{…}

R:{…} R:{…}

49

Power Management Approach

Coarse-grained app-controlled communication scheduling

1

2

3

4

5

Mote ID

time

Epoch (10s -100s of seconds)

2-4s Waking Period

… zzz … … zzz …

50

Time Synchronization

• All messages include a 5 byte time stamp indicating system time in ms– Synchronize (e.g. set system time to timestamp) with

• Any message from parent• Any new query message (even if not from parent)

– Punt on multiple queries– Timestamps written just after preamble is xmitted

• All nodes agree that the waking period begins when (system time % epoch dur = 0)– And lasts for WAKING_PERIOD ms

• Adjustment of clock happens by changing duration of sleep cycle, not wake cycle.

51

Extending TinyDB

• Why extending TinyDB?– New sensors attributes– New control/actuation commands– New data processing logic

aggregates– New events

• Analogous to concepts in object-relational databases

52

Adding Attributes

• Types of attributes– Sensor attributes: raw or cooked

sensor readings– Introspective attributes: parent,

voltage, ram usage, etc.– Constant attributes: constant values

that can be statically or dynamically assigned to a mote, e.g., nodeid, location, etc.

53

Adding Attributes (cont)

• Interfaces provided by Attr component– StdControl: init, start, stop– AttrRegister

• command registerAttr(name, type, len)• event getAttr(name, resultBuf, errorPtr)• event setAttr(name, val)• command getAttrDone(name, resultBuf, error)

– AttrUse• command startAttr(attr)• event startAttrDone(attr)• command getAttrValue(name, resultBuf, errorPtr)• event getAttrDone(name, resultBuf, error)• command setAttrValue(name, val)

54

Adding Attributes (cont)

• Steps to adding attributes to TinyDB1) Create attribute nesC components2) Wire new attribute components to

TinyDBAttr configuration 3) Reprogram TinyDB motes4) Add new attribute entries to catalog.xml

• Constant attributes can be added on the fly through TinyDB GUI

55

Adding Aggregates

• Step 1: wire new nesC components

56

Adding Aggregates (cont)

• Step 2: add entry to catalog.xml<aggregate>

<name>AVG</name>

<id>5</id>

<temporal>false</temporal>

<readerClass>net.tinyos.tinydb.AverageClass</readerClass>

</aggregate>

• Step 3 (optional): implement reader class in Java– a reader class interprets and finalizes aggregate state

received from the mote network, returns final result as a string for display.

57

TinyDB Status

• Latest released with TinyOS 1.1 (9/03)– Install the task-tinydb package in TinyOS 1.1

distribution– First release in TinyOS 1.0 (9/02)– Widely used by research groups as well as industry pilot

projects

• Successful deployments in Intel Berkeley Lab and redwood trees at UC Botanical Garden– Largest deployment: ~80 weather station nodes– Network longevity: 4-5 months

58

The Redwood Tree Deployment

• Redwood Grove in UC Botanical Garden, Berkeley

• Collect dense sensor readings to monitor climatic variations across– altitudes,– angles,– time,– forest locations, etc.

• Versus sporadic monitoring points with 30lb loggers!

• Current focus: study how dense sensor data affect predictions of conventional tree-growth models

59

Humidity vs. Time

35

45

55

65

75

85

95

Rel Humidity (%)

101 104 109 110 111

Data from Redwoods

36m

33m: 111

32m: 110

30m: 109,108,107

20m: 106,105,104

10m: 103, 102, 101

Temperature vs. Time

8

13

18

23

28

33

7/7/039:40

7/7/0313:11

7/7/0316:43

7/7/0320:15

7/7/0323:46

7/8/033:18

7/8/036:50

7/8/0310:21

7/8/0313:53

7/8/0317:25

7/8/0320:56

7/9/030:28

7/9/034:00

7/9/037:31

7/9/0311:03

Date

Temperature (C)

60

TinyDB Roadmap (near term)

• Support for high frequency sampling– Equipment vibration monitoring, structural

monitoring, etc.– Store and forward– Bulk reliable data transfer– Scheduling of communications

• Port to Intel Mote• Deployment in Intel Fab equipment monitoring

application and the Golden Gate Bridge monitoring application

61

For more information

• http://berkeley.intel-research.net/tinydb or http://triplerock.cs.bekeley.edu/tinydb

62

Part 3

Database Research Issues in Sensor Networks

63

Sensor Network Research

• Very active research area– Can’t summarize it all

• Focus: database-relevant research topics– Some outside of Berkeley– Other topics that are itching to be scratched– But, some bias towards work that we find

compelling

64

Topics

• In-network aggregation• Acquisitional Query Processing• Heterogeneity• Intermittent Connectivity• In-network Storage• Statistics-based summarization and

sampling• In-network Joins• Adaptivity and Sensor Networks• Multiple Queries

65

Topics



66

Tiny Aggregation (TAG)

• In-network processing of aggregates– Common data analysis operation

• Aka gather operation or reduction in || programming

– Communication reducing• Operator dependent benefit

– Across nodes during same epoch

• Exploit query semantics to improve efficiency!

Madden, Franklin, Hellerstein, Hong. Tiny AGgregation (TAG), OSDI 2002.

67

Basic Aggregation

• In each epoch:– Each node samples local sensors once– Generates partial state record (PSR)

• local readings • readings from children

– Outputs PSR during assigned comm. interval

• At end of epoch, PSR for whole network output at root

• New result on each successive epoch

• Extras:– Predicate-based partitioning via GROUP BY

1

2 3

4

5

68

Illustration: Aggregation

1 2 3 4 5

4 1

3

2

1

4

1

2 3

4

5

1

Sensor #

Inte

rval #

Interval 4SELECT COUNT(*) FROM sensors

Epoch

69


1 2 3 4 5

4 1

3 2

2

1

4

1

2 3

4

5

2

Sensor #


Inte

rval #

70


1 2 3 4 5

4 1

3 2

2 1 3

1

4

1

2 3

4

5

31

Sensor #


Inte

rval #

71


1 2 3 4 5

4 1

3 2

2 1 3

1 5

4

1

2 3

4

5

5

Sensor #

SELECT COUNT(*) FROM sensors Interval 1

Inte

rval #

72


1 2 3 4 5

4 1

3 2

2 1 3

1 5

4 1

1

2 3

4

5

1

Sensor #

SELECT COUNT(*) FROM sensors Interval 4

Inte

rval #

73

Aggregation Framework

• As in extensible databases, TinyDB supports any aggregation function conforming to:

Aggn={finit, fmerge, fevaluate}

Finit {a0} <a0>

Fmerge {<a1>,<a2>} <a12>

Fevaluate {<a1>} aggregate value

Example: AverageAVGinit {v} <v,1>

AVGmerge {<S1, C1>, <S2, C2>} < S1 + S2 , C1 + C2>

AVGevaluate{<S, C>} S/C

Partial State Record (PSR)

Restriction: Merge associative, commutative

74

Property Examples Affects

Partial State MEDIAN : unbounded, MAX : 1 record

Effectiveness of TAG

Monotonicity COUNT : monotonicAVG : non-monotonic

Hypothesis Testing, Snooping

Exemplary vs. Summary

MAX : exemplaryCOUNT: summary

Applicability of Sampling, Effect of Loss

Duplicate Sensitivity

MIN : dup. insensitive,AVG : dup. sensitive

Routing Redundancy

Taxonomy of Aggregates

• TAG insight: classify aggregates according to various functional properties– Yields a general set of optimizations that can automatically be

applied

Drives an API!

75

Use Multiple Parents

• Use graph structure – Increase delivery probability with no communication

overhead

• For duplicate insensitive aggregates, or• Aggs expressible as sum of parts

– Send (part of) aggregate to all parents• In just one message, via multicast

– Assuming independence, decreases variance

SELECT COUNT(*)

A

B C

R

A

B C

c

R

P(link xmit successful) = p

P(success from A->R) = p2

E(cnt) = c * p2

Var(cnt) = c2 * p2 * (1 – p2) V

# of parents = n

E(cnt) = n * (c/n * p2)

Var(cnt) = n * (c/n)2 * p2 * (1 – p2) = V/n

A

B C

c/n c/n

R

n = 2

76

Multiple Parents Results

• Better than previous analysis expected!

• Losses aren’t independent!

• Insight: spreads data over many links

Benefit of Result Splitting (COUNT query)

0

200

400

600

800

1000

1200

1400

(2500 nodes, lossy radio model, 6 parents per node)

Avg. COUNT

Splitting

No Splitting

Critical Link!

No Splitting With Splitting

77

Acquisitional Query Processing (ACQP)

• TinyDB acquires AND processes data

– Could generate an infinite number of samples

• An acqusitional query processor controls

– when,

– where,

– and with what frequency data is collected!

• Versus traditional systems where data is provided a priori

Madden, Franklin, Hellerstein, and Hong. The Design of An Acqusitional Query Processor. SIGMOD, 2003.

78

ACQP: What’s Different?• How should the query be processed?

– Sampling as a first class operation

• How does the user control acquisition?– Rates or lifetimes– Event-based triggers

• Which nodes have relevant data?– Index-like data structures

• Which samples should be transmitted?– Prioritization, summary, and rate control

79

• E(sampling mag) >> E(sampling light)

1500 uJ vs. 90 uJ

Operator Ordering: Interleave Sampling + Selection

SELECT light, magFROM sensorsWHERE pred1(mag)AND pred2(light)EPOCH DURATION 1s

(pred1)

(pred2)

mag

light

(pred1)

(pred2)

mag

light

(pred1)

(pred2)

mag light

Traditional DBMS

ACQP

At 1 sample / sec, total power savings could be as much as 3.5mW Comparable to processor!

Correct orderingCorrect ordering(unless pred1 is (unless pred1 is very very selective selective

and pred2 is not):and pred2 is not):

Cheap

Costly

80

Exemplary Aggregate Pushdown

SELECT WINMAX(light,8s,8s)FROM sensorsWHERE mag > xEPOCH DURATION 1s

• Novel, general pushdown technique

• Mag sampling is the most expensive operation!

WINMAX

(mag>x)

mag light

Traditional DBMS

light

mag

(mag>x)

WINMAX

(light > MAX)

ACQP

81

Topics

• In-network aggregation• Acquisitional Query Processing• Heterogeneity• Intermittent Connectivity• In-network Storage• Statistics-based summarization and sampling• In-network Joins• Adaptivity and Sensor Networks• Multiple Queries

82

Heterogeneous Sensor Networks

• Leverage small numbers of high-end nodes to benefit large numbers of inexpensive nodes

• Still must be transparent and ad-hoc• Key to scalability of sensor networks• Interesting heterogeneities

– Energy: battery vs. outlet power– Link bandwidth: Chipcon vs. 802.11x– Computing and storage: ATMega128 vs.

Xscale– Pre-computed results– Sensing nodes vs. QP nodes

83

Computing Heterogeneity with TinyDB

• Separate query processing from sensing– Provide query processing on a small number of nodes– Attract packets to query processors based on “service

value”• Compare the total energy consumption of the

network

• No aggregation• All aggregation• Opportunistic aggregation• HSN proactive

aggregation

Mark Yarvis and York Liu, Intel’s Heterogeneous Sensor

Network Project, ftp://download.intel.com/research/people/HSN_IR_Day_Poster_03.pdf.

84

5x7 TinyDB/HSN Mica2 Testbed

85

Data Packet SavingData Packet Saving

-50.00%

-45.00%

-40.00%

-35.00%

-30.00%

-25.00%

-20.00%

-15.00%

-10.00%

-5.00%

0.00%

1 2 3 4 5 6 All (35)

Number of Aggregator

% Change in Data Packet Count

Data Packet Saving - Aggregator Placement

-50.00%

-45.00%

-40.00%

-35.00%

-30.00%

-25.00%

-20.00%

-15.00%

-10.00%

-5.00%

0.00%

25 27 29 31 All (35)

Aggregator Location

% Change in Data Packet Counnt

• How many aggregators are desired?

• Does placement matter?

11% aggregators achieve 72% of max

data reduction

Optimal placement 2/3 distance from sink.

86

Occasionally Connected Sensornets

TinyDB QPTinyDB QP

TinyDB Server

GTWY

Mobile GTWYMobile GTWY

TinyDB QP

Mobile GTWY

GTWY

internet

GTWY

87

Occasionally Connected Sensornets Challenges

• Networking support– Tradeoff between reliability, power

consumption and delay– Data custody transfer: duplicates?– Load shedding– Routing of mobile gateways

• Query processing– Operation placement: in-network vs. on mobile

gateways– Proactive pre-computation and data movement

• Tight interaction between networking and QP

Fall, Hong and Madden, Custody Transfer for Reliable Delivery in Delay Tolerant Networks, http://www.intel-research.net/Publications/Berkeley/081220030852_157.pdf.

88

Distributed In-network Storage

• Collectively, sensornets have large amounts of in-network storage

• Good for in-network consumption or caching

• Challenges– Distributed indexing for fast query

dissemination– Resilience to node or link failures– Graceful adaptation to data skews– Minimizing index insertion/maintenance cost

89

Example: DIM• Functionality

– Efficient range query for multidimensional data.

• Approaches– Divide sensor field into

bins.– Locality preserving

mapping from m-d space to geographic locations.

– Use geographic routing such as GPSR.

• Assumptions– Nodes know their

locations and network boundary

– No node mobility

E2= <0.6, 0.7>E1 = <0.7, 0.8>

Q1=<.5-.7, .5-1>

Xin Li, Young Jin Kim, Ramesh Govindan and Wei Hong, Distributed Index for Multi-dimentional Data (DIM) in Sensor Networks, SenSys 2003.

90

Statistical Techniques

• Approximations, summaries, and sampling based on statistics and statistical models

• Applications:– Limited bandwidth and large number of

nodes -> data reduction– Lossiness -> predictive modeling– Uncertainty -> tracking correlations and

changes over time– Physical models -> improved query

answering

91

Correlated Attributes

• Data in sensor networks is correlated; e.g.,– Temperature and voltage– Temperature and light– Temperature and humidity– Temperature and time of day– etc.

92

IDSQ

• Idea: task sensors in order of best improvement to estimate of some value:– Choose leader(s)

• Suppress subordinates• Task subordinates, one at a time

– Until some measure of goodness (error bound) is met

» E.g. “Mahalanobis Distance” -- Accounts for correlations in axes, tends to favor minimizing principal axis

See “Scalable Information-Driven Sensor Querying and Routing for ad hoc Heterogeneous Sensor Networks.” Chu, Haussecker and Zhao. Xerox TR P2001-10113. May, 2001.

93

Model location estimate as a point with 2-dimensional Gaussian uncertainty.

Graphical Representation

Principal Axis

S1

Residual 1

Preferred because it reduces error along principal axis

Residual 2 S2

Area of residuals is equal

94

MQSN: Model-based Probabilistic Querying over

Sensor Networks

Query ProcessorModel

1

23

5

6

4

7

8

9

Joint work with Amol Desphande, Carlos Guestrin,

and Joe Hellerstein

95


Sensor Networks


1

23

5

6

4

7

8

9

Probabilistic Queryselect NodeID, Temp ± 0.1Cwhere NodeID in [1..9] with conf(0.95)

Consult

Model

Observation Plan[Temp, 3], [Temp, 9]

96


Sensor Networks


1

23

5

6

4

7

8

9

Observation Plan[Temp, 3], [Temp, 9]

Probabilistic Queryselect NodeID, Temp ± 0.1Cwhere NodeID in [1..9] with conf(0.95)

Consult

Model

97


Sensor Networks


1

23

5

6

4

7

8

9

Data[Temp, 3] = …, [Temp, 9] = …

Query Results

10

15

20

25

30

1 2 3 4

Node ID

Temperature

Update

Model

98

Challenges

• What kind of models to use ?• Optimization problem:

– Given a model and a query, find the best set of attributes to observe

– Cost not easy to measure• Non-uniform network communication costs• Changing network topologies

– Large plan space• Might be cheaper to observe attributes not in query

– e.g. Voltage instead of Temperature• Conditional Plans:

– Change the observation plan based on observed values

99

MQSN: Current Prototype

• Multi-variate Gaussian Models– Kalman Filters to capture correlations across time

• Handles:– Range predicate queries

• sensor value within [x,y], w/ confidence– Value queries

• sensor value = x, w/in epsilon, w/ confidence– Simple aggregate queries

• AVG(sensor value) n, w/in epsilon, w/confidence

• Uses a greedy algorithm to choose the observation plan

100

In-Net Regression

• Linear regression : simple way to predict future values, identify outliers

• Regression can be across local or remote values, multiple dimensions, or with high degree polynomials– E.g., node A readings vs. node B’s– Or, location (X,Y), versus temperature

E.g., over many nodes

X vs Y w/ Curve Fit

y = 0.9703x - 0.0067

R2 = 0.947

0

2

4

6

8

10

12

1 3 5 7 9Guestrin, Thibaux, Bodik, Paskin, Madden. “Distributed Regression: an Efficient

Framework for Modeling Sensor Network Data .” Under submission.

101

In-Net Regression (Continued)

• Problem: may require data from all sensors to build model

• Solution: partition sensors into overlapping “kernels” that influence each other– Run regression in each kernel

• Requiring just local communication

– Blend data between kernels– Requires some clever matrix manipulation

• End result: regressed model at every node– Useful in failure detection, missing value

estimation

102

Exploiting Correlations in Query Processing

• Simple idea: – Given predicate P(A) over expensive attribute A– Replace it with P’ over cheap attribute A’ such that P’

evaluates to P – Problem: unless A and A’ are perfectly correlated, P’ ≠ P

for all time• So we could incorrectly accept or reject some readings

• Alternative: use correlations to improve selectivity estimates in query optimization– Construct conditional plans that vary predicate order

based on prior observations

103

Exploiting Correlations (Cont.)

• Insight: by observing a (cheap and correlated) variable not involved in the query, it may be possible to improve query performance – Improves estimates of selectivities

• Use conditional plans• Example

Light > 100 Lux

Temp < 20° C

Cost = 100Selectivity = .5


Expected Cost = 150

Light > 100 Lux

Temp < 20° C



Expected Cost = 150

Light > 100 Lux

Temp < 20° C



Expected Cost = 110

Light > 100 Lux

Temp < 20° C



Expected Cost = 110

Time in [6pm, 6am]

T

F

104

In-Network Join Strategies

• Types of joins: – non-sensor -> sensor– sensor -> sensor

• Optimization questions:– Should the join be pushed down?– If so, where should it be placed?– What if a join table exceeds the

memory available on one node?

105

Choosing Where to Place Operators

• Idea : choose a “join node” to run the operator

• Over time, explore other candidate placements– Nodes advertise data rates to their neighbors– Neighbors compute expected cost of running the

join based on these rates– Neighbors advertise costs– Current join node selects a new, lower cost node

Bonfils + Bonnet, Adaptive and Decentralized Operator Placement for In-Network QueryProcessing IPSN 2003.

106

Topics



107

Adaptivity In Sensor Networks

• Queries are long running• Selectivities change

– E.g. night vs day

• Network load and available energy vary• All suggest that some adaptivity is needed

– Of data rates or granularity of aggregation when optimizing for lifetimes

– Of operator orderings or placements when selectivities change (c.f., conditional plans for correlations)

• As far as we know, this is an open problem!

108

Multiple Queries and Work Sharing

• As sensornets evolve, users will run many queries simultaneously– E.g., traffic monitoring

• Likely that queries will be similar– But have different end points, parameters,

etc

• Would like to share processing, routing as much as possible

• But how? Again, an open problem.

109

Concluding Remarks

• Sensor networks are an exciting emerging technology, with a wide variety of applications

• Many research challenges in all areas of computer science– Database community included– Some agreement that a declarative interface is right

• TinyDB and other early work are an important first step

• But there’s lots more to be done!

tinydb tutorial

Documents

networks of sensors

types of sensors sensors

lifetime power

berkeley motes

tinysql power

power hungry

network research

kbit radio