This image cannot currently be displayed.

Massive Data Computing Research Center

Application-Aware Data Collection in Wireless Sensor Networks

Fang Xiaolin, Gao Hong, Li JianzhongHarbin Institute of Technology

Li YingshuGeorgia State University

2Massive Data Computing Research Center

Outline

Existing workOur problemAn approximation algorithmA special instanceSimulations

3Massive Data Computing Research Center

Existing work

Multi-application based data collectionData sharing [9]Sample as less data as possible

Data point

Existing work studies data point sampling

4Massive Data Computing Research Center

Our problem

Multi-application based data collectionSample data for a continuous interval

Acoustic, video information [10], [11] Vibration measurement [12], [13], [14]Speed information [15]

Sample an interval

5Massive Data Computing Research Center

Problem definition

Given a set of n tasks T , each task Tiis denoted as Ti = <bi,ei,li>

bi: beginning timeei: end timeli: data sampling interval length

Find a continuous sub-interval Ii for each task Ti so that

6Massive Data Computing Research Center

Problem complexity

Non-linear non-convex optimization problem

Nonlinear integer programming problem, If bi,ei,li are regarded as integers

7Massive Data Computing Research Center

Greedy algorithm overview

1. Sort by end times 2. Find task set P overlap with first 3. Find solution for P, and Remove 4. Back to step 2

8Massive Data Computing Research Center

Greedy algorithm overview

1. Sort by end times 2. Find task set P overlap with first 3. Find solution for P, and Remove 4. Back to step 2

9Massive Data Computing Research Center

Greedy algorithm overview

1. Sort by end times 2. Find task set P overlap with first 3. Find solution for P, and Remove 4. Back to step 2

10Massive Data Computing Research Center

Greedy algorithm overview

1. Sort by end times 2. Find task set P overlap with first 3. Find solution for P, and Remove 4. Back to step 2

11Massive Data Computing Research Center

Find solution for P

Compute [s,e] for the tasks overlap with each other

s=5 e=14

[s,e]=[5,14]

12Massive Data Computing Research Center

Approximation algorithm analysis

Approximation ratio is 2Time complexity is

13Massive Data Computing Research Center

A special instance

General problem Ti = <bi,ei,li>Special instance Ti = <bi,ei,l>The data lengths are the same

Can be solved in O(n2)

14Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

Does not affect the result

15Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

16Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

Computing x(i,j), then x(1,n) is the result[3,7]

[5,9][12,16]

x(i,j) [s,e]x(1,1) [3,7]x(2,2) [5,9]x(3,3) [12,16]x(1,2) [3,8]x(2,3) [5,10]

17Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

Computing x(i,j), then x(1,n) is the result[3,8]

x(i,j) [s,e]x(1,1) [3,7]x(2,2) [5,9]x(3,3) [12,16]x(1,2) [3,8]x(2,3) [5,10]

18Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

Computing x(i,j), then x(1,n) is the result

[5,10] x(i,j) [s,e]x(1,1) [3,7]x(2,2) [5,9]x(3,3) [12,16]x(1,2) [3,8]x(2,3) [5,10]

19Massive Data Computing Research Center

Algorithm overview

1. Sort by the end times 2. Remove tasks cover other tasks 3. Dynamic programming

Computing x(i,j), then x(1,n) is the result[3,10]

x(i,j) [s,e]x(1,1) [3,7]x(2,2) [5,9]x(3,3) [12,16]x(1,2) [3,8]x(2,3) [5,10]x(1,3) =

x(1,1) U x(2,3)+

x(1,2) U x(3,3)+min

20Massive Data Computing Research Center

Simulation

TossimFour casesTasks are from multi-applicationsEach application consists of periodical

tasks

21Massive Data Computing Research Center

Simulation

short sampling interval lengths

22Massive Data Computing Research Center

Simulation

longer sampling interval lengths

23Massive Data Computing Research Center

Simulation

Different Window size

24Massive Data Computing Research Center

Simulation

Data loss

25Massive Data Computing Research Center