Combs, Needles, Haystacks: Balancing Push and Pull for Discovery in Large Scale Sensor Networks Xin...

Combs, Needles, Haystacks: Balancing Push and Pull for Discovery in Large

Scale Sensor Networks

Xin LiuDepartment of Computer ScienceUniversity of California

Qingfeng Huang and Ying ZhangPalo Alto Research Center (PARC) Inc.

SenSys 2004

Presenter ： Ruey-Chang Chang

2

Outline

Introduction The combing strategy Adaptive comb-needle strategy Simulation Conclusion

3

Push-based

4

Pull based

5

Introduction

Push-pull strategies for discovery– Push-based

The push-based strategy is efficient when there are many sinks constantly in need of the information

A lot of broadcast bandwidth is wasted– Pull-based

The pull-based strategy is relative more efficient than the push-based strategy when the frequency of query is relatively low compared to the frequency of the interested event

– Hybrid (a comb-needle strategies) Combine the advantages of both push and pull strategies

6

Hybrid (a comb-needle strategies)

7

The combing strategy

In the comb-needle model– Each sensor node pushes its data to a certain

neighborhood and the query is disseminated only to a subset of the network

– The query process builds a routing structure dynamically that resembles a comb

– The sensor node push the data duplication structure like a needle

8

The combing strategy

l

s

9

The total cost per query

Cost of Query dissemination

Cost of Query response per each node

Cost of Query response

Cost of data push

fe:the arrival frequency of discovery queriesfq:the arrival frequency of relevant events

10

The minimal cost

s=2l+1

l

s

sink

sensor

l

s

sink

sensor

11

Routing protocol

Constrained Geographical Flooding– Whenever a new packet arrives, each node will decide if it

should rebroadcast the packet according to the geographical constraints W

W

12

fqfe(Global-pull-local-push)

push

pullsink

sensor

13

fq=fe

push

push

pull

14

fqfe(Global-push-local-pull)

This paper focuses on fqfe

push

pull

15

Adaptive comb-needle strategy

The query and event frequencies may be time-varying ,and thus a good query strategy should adapt to such change

fe:the arrival frequency of discovery queries

fq:the probability that a query is generated in a time slot

fd:the probability that a sensor node detects an event in a time slot

16

Simulation1

fq/fe=0.1(fe= 1 packet/second,fq= 0.1p/s) l={0,1,2,3} s={1,3,5,7} Node:? Simulation:? Topology:grid

17

Energy consumption(fq/fe=0.1)

L=1 S=3

18

Energy consumption(fq/fe=1)

L=2 S=5

19

Energy consumption

20

Simulation2

3000 time slots 20*20 grid

21

The query and the event frequency

22

Adaptive comb vs. ideal comb

23

Conclusion

The comb-needle model– A simple yet efficient data discovery scheme for

supporting queries– A substrate for study the benefit of balancing

push and pull in data gathering and dissemination– Covers a spectrum of the push and pull schemes– An adaptive comb-needle strategy for cases

where the utilization patterns and environmental activity frequency are time-frequency