Query Processing in WSN: The

brainybootsMobile - Wireless

Nov 21, 2013 (3 years and 6 months ago)

60 views

Baljeet Malhotra


Supervisors

Ioanis Nikolaidis

Mario A. Nascimento


Communication Networks and Database Systems

Computing Science Department





Graduate Student’s Workshop on Network’s Research, Nov 17, 2009


Query Processing in WSN: The
Dilemma of Suppressions and Failures

In
-
Network Query Processing in WSN

Query

Result

In
-
network Processing

Decision Support system

Background


Queries


Variations of Top
-
k (Min, Max, Avg.)


Nearest Neighbor


Join queries


Techniques


Exploit the wireless broadcast nature of WSN


Aggregation, Pruning, Filtering ….


Schedules


Sensors should not be in ‘idle listening mode”


Contentious our slotted time


Failures


Unreliable communication


?

Top
-
k
Query

:
Problem Definition


Given a set of
N
sensors,
S =
{
s
i
:

i =1, 2,….., N
}



S
p,j

be the set of sensors that produced the
p
th

highest
value during the
j
th

round


v
(
S
p,j
) be the
p
th

highest value



The problem is to find:


Top
-
k values:
D
i

=
{
v
(
S
p,j
)

:

p =1, 2,….., k
}


Top
-
k sensors:
S
p,j


Top
-
k
Query Processing

A

Sink = Root

Shortest Path Tree (SPT)

Convergecast + Aggregation = TAG*

A

Sink = Root

*TAG, Madden et. al., OSDI’02, 2002

Convergecast + Aggregation

A

Sink = Root

Total # of messages used = n
-
1

Convergecast + Aggregation + Filtering

A

Sink = Root

τ (threshold) = min of top
-
k

Threshold (
τ
) Broadcast

A

Sink = Root

Convergecast + Aggregation + Filtering

A

Sink = Root

send iff
v(
s
i
) ≥
τ
; or if top
-
k
node

Broadcast via SPT

A

Sink = Root

Total # of messages used = 8

Broadcast via DST

A

Sink = Root

Dominating Set Tree (DST)

Total # of messages used = 4

Some Results


Varying
k









Tree topology makes a difference


EXTOK performs better on both synthetic and Intel
dataset


Synthetic data

Real data

Convergecast Scheduling

A

Sink = Root

TDMA slots => Precedence Constraints

Convergecast Scheduling

A

Sink = Root



TDMA slots => Conflict Free + Precedence Constraints

1

2

3

3

4

5

5

6

8

7

Some Results


Varying
Ψ

(node density)










SDA, Chen et. al, LNCS, vol. 3799, 2005


SAS, Wan, et. al., MOBIHOC, 2009


DAS, B. Yu et. al., INFOCOM, 2009


PAS, X. Yu et. al., SSDB, 2007

Synthetic data

Real data

Convergecast Scheduling + Filtering

A

Sink = Root

send iff
v(
s
i
) ≥
τ
; or if top
-
k
node

1

2

3

3

4

5

5

6

8

7

Convergecast Scheduling + Filtering + Failure

A

Sink = Root

send iff
v(
s
i
) ≥
τ
; or if top
-
k
node

1

2

3

3

4

5

5

6

8

7

Convergecast Scheduling + Failure Recovery

A

Sink = Root

1

2

3

3

4

5

5

6

8

7



Some Results

Application Perspective

Just Brodcast

Conclusions


Infrastructure


Clusters and Routing Trees


Dominating Sets


Techniques


Aggregation, Pruning, and Filtering


How to use them in the best possible way for a particular problem ?


Can we use one single infrastructure for every thing ?


Schedules


Every task must be done in a systematic fashion while minimizing the
response time


Failures


What happens when parts of our infrastructure breaks down ?


How and When to fix these problems ?

Acknowledgements


This research is partially supported by NSERC,
i
-
Core,
FGSR, VLDB, GSA, and Walter John Scholarship.




Contact
baljeet@cs.ualberta.ca

for more details.



THANK YOU !