An adaptive framework of multiple schemes for

event and query distribution in wireless

sensor networks

Vincent Tam, Keng-Teck Ma, and King-Shan Lui

IEEE Consumer Communications and Networking Conference (CCNC’07)

Outline Introduction Adaptive framework using multiple

schemes Simulation Conclusion

Approaches of getting information Push-based approach

Floods to all nodes No query

Pull-based approach Stores locally Floods the network with a query

Data-centric storage (DCS) Depends on the type information

Computed using a globally hash function

An important issue of DCS Where to keep event information

Keeping information in the center Increase the workload of the nodes

Varying query/event ratio Affect the performance of various data

dissemination schemes An adaptive framework

Flexibly switch from one scheme to another Based on the results evaluated by cost

The cost analysis of local storage

The total message cost for Local Storage (LS):

nDnQ q

n: the number of nodes

Q: the number of queries

Dq:the number of events for queries

Dtotal: the total number of events

The message overheads of flooding

to n nodesThe average message cost to answer all the

queries

The cost analysis of data-centric storage

The total message cost for Data-Centric Storage (DCS):

nDnDnQ qtotal

n: the number of nodes

Q: the number of queries

Dq:the number of events for queries

Dtotal: the total number of events

The average message cost for routing the queries to hashed

nodes

The average message cost associated with the storage of events

being detected

The average message cost to answer all the

queries

Cost AnalysisnQnDnQ total

)1/(1/ nDQ total

implying

)1/(1/ nDQ total

GHT outperforms the LS with a small overall message

overheads

LS excelled the GHT with a small overall message

overheads

In many real-world applications of wireless sensor networks, such of (query : event) can be dynamically

changing over time

Adaptive framework using multiple schemes Initially, framework starts with LS

Relatively few events detected in the beginning

When event is detected Stores in detecting node

When a query is generated Route to home node

(50, 89)

SinkGetGet(“(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion

”)”)

Adaptive framework using multiple schemes Upon a query is received

Compute the cost of LS and GHT LS is higher than GHT

From LS to GHT GHT is higher than LS

From GHT to LS

Any node with the locally stored event Reply to the home node

Home node update the count for events being detected

Relay the event back to query node

)1/(1/ nDQ total

Switching from LS to GHT

Sensor nodeSensor node

home nodehome node

LocationLocation

Mapping Mapping location for lionlocation for lion

Firstly setting the GHT flag in the next arriving query as still under the

LS mode

SinkGetGet(“(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion

”)”)

Each node reset itself to the GHT mode

After switching from LS to GHT

Sensor nodeSensor node

home nodehome node

LocationLocation

Mapping Mapping location for lionlocation for lion

Put(“Put(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion”)”)

Sink

GetGet(“(“lionlion”) ”) (( 5050,8,899)=Hash(“)=Hash(“lionlion””

))

)1/(1/ nDQ total

Switching from GHT to LS

Sensor nodeSensor node

home nodehome node

LocationLocation

Mapping Mapping location for lionlocation for lion

Firstly switch itself to LS mode after answering

the query

Put(“Put(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion”)”)

Set LS flag embedded in the ACK packets for all the future events being

detected

)1/(1/ nDQ total

Switching from GHT to LS

Sensor nodeSensor node

home nodehome node

LocationLocation

Mapping Mapping location for lionlocation for lion

SinkGetGet(“(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion

”)”)

Each node reset itself to the LS mode

After switching from GHT to LS

Sensor nodeSensor node

home nodehome node

LocationLocation

Mapping Mapping location for lionlocation for lion

Put(“Put(“lionlion”, data) ”, data) (( 5050,8,899)=Hash(“)=Hash(“lionlion”)”)

Sink

GetGet(“(“lionlion”) ”) (( 5050,8,899)=Hash(“)=Hash(“lionlion””

))

A list of the recently events and queries Each home node maintain a list of the recently

received events and queries Compute the number of events or queries

Each entry is time-stamped A sliding window filter out those expired events or

queries

Sliding window size (time)

Past data Future data

Simulation

Simulation

Simulation

Simulation

Conclusion The query-to event ration

Can be changing over time

An adaptive frame work Flexibly switch from one scheme to another Based on the message cost

To minimize the overall message overheads