Nonblocking GHT for datacentric storage

1
Non-blocking GHT for data-centric storage

• Kwangjae Yeo and Jongyoon Choi

2
Table of contents

• Problem Statement
• Related Works
• Design Principle
• Architecture
• Operations
• Simulation
• Conclusion

3
Problem Statement

• Different architecture is required for different
  data type

Statistic Data

• Average, Median, Maximum, Minimum
• Possible to generated relevant data instead of
  read individual node
• Predictable in terms of examining time

Detect Events

• Transient event Animal sightings
• Need to access each individual node
• Unpredictable time to event occurring

• What we want to do is

• Effectively detect a transient event
• Efficiently store a transient event
• Explicitly query a transient event

in Location Aware Environments
4
Related Works 1

• Query processing has been studied to decentralize
  operation

G H T

• data is hashed by event name to a location
• flooding is required to announce event for
  hashing
• hash function is still vague ex) definition of
  key

Directed Diffusion

• diffusion to select empirically good paths
• diverse propagation model
• not related to querying events.

R E E D

• provide efficient query with complex predicates
• supports join in-network like standard SQL
• relatively heavy for tiny sensor and requires
  the coordinator

5
Related Works 2

• Hierarchical architecture is studied for
  efficiency

D I F S

• use quad tree to form a search tree
• use of all nodes for data store

DIMENSION

• uses grid topology model and multi-resolution
  data storage
• adaptation of correlations in sensor data

D I M

• focused on range query problem with associated
  zones
• definition of a zone is independent of the
  actual location of node

6
Design Principle Event Detection

• Grid-Frisbee minimize the power usage by using
  an active zone

• each sensor has operating mode and power saving
  mode.
• leaf grid head has responsibility of sentry
• If grid head detects an event, it sends signal
  to all eight neighbor grids.

• The entire field is divided as four sub grids
• Each sub grid also decomposed by four
• Definition Leaf grid, Root grid

7
Design Principle Leader Election

• Each leaf grid should have 4 special nodes

• Main backup grid head and Main backup DCS
• These are selected by leader election algorithm
• Main grid head compete with other three grid
  heads for parent head role
• It propagates to root head role

8
Design Principle Event Hierarchy

• Each of grid head holds a Grid Event Table (GET)

• Each leaf grid head stores only the event
  history of its own
• Parent has four entries for each immediate child
  grid
• Decide its own history by AND operation
• The entry of root grid is zero, if no event
  happens.

9
Design Principle Data Structure

• Each DCS holds the Event History List (EHL)

• DCS stores the most recent event to the head of
  list

• Each grid head has a grid geographic hash table
  (Grid GHT)

• The result of event query is the geographic
  coordinates of grid
• Each node can route to grid with GPSR

10
Architecture

• Efficiency by balancing the load across the
  sensor field

11
Operations
Initialization

• Decide a role by leader election and Construct
  an event hierarchy

Sensing

• Localize active zone by grid Frisbee model

Storing

• Store event information to DCS and Report it to
  Grid Event Table

Querying

• Forward query to root and start to search using
  Grid Event Table

Routing

• Return grid id where event happens and route
  with GPSR

Refreshing

• Re-do election if requested

12
Simulation Configuration

• NS2 Configuration

Node location setup

• Number of awakened nodes 256, 128, 64, 16
• Area 80m 80 m
• Duration 100 Sec
• Transmission radio range 9m

Detect Events
Protocol IEEE 802.15.4

• Wireless MAC and PHY layer specifications for
  Low-rate Wireless Personal Area Networks
  (LR-WPANs)
• Very low power consumption
• Simple but flexible protocol
• Energy detection (ED)
• Link quality indication (LQI)
• Allocation of guaranteed time slots (GTSs)
• CSMA/CA
• Star topology/Peer-to-peer topology

13
Simulation result

• The number of transmissions increases linearly

• The number of transmission increases
  exponentially in GHT
• Because of flooding in registering event

• The number of nodes increases exponentially
• However, the number of transmission increases
  linearly
• It includes the number of transmission of MAC
  Layer of 802.15.4 protocol.

14
Conclusion

• Method of multi-level hierarchical aggregation
  of transitional events ? Grid Frisbee, Grid
  Event Table, Grid Leader Election, Event
  History List (EHL), Grid GHT
• Iterative operation cycle Initialization ?
  Sensing ? Storing ? Querying ? Routing ?
  Refreshing

Experience from Result

• Hierarchical approach works for transitional
  event ? Leverage work load all over the
  network ? Extend the minimum time to power
  down of node (not average)

• Future study ? Need to study the case without
  location information ? Join query with other
  event and time information

15
References
1 Sylvia Ratnasamy, GHT A Geographic Hash
Table for Data-Centric Storage, ACM WSNA02 2
B. Greenstein, DIFS A distributed index for
features in sensor networks. 1st IEEE workshop
on sensor networks protocols and applications,
2003. 3 Chalermek Intanagonwiwat, Directed
diffusion A scalable and robust communication
paradigm for sensor networks, ACM/IEEE MobiCOM
'00 4 Deepak Ganesan, DIMENSIONS why do we
need a new data handling architecture for sensor
networks. ACM SIGCOMM 02. 5 Alberto Cerpa,
Habitat monitoring application driver for
wireless communications technology , ACM SIGCOMM
Computer Communication Review, Volume 31 , Issue
2 supplement (April 2001) 8 Xin Li,
Multi-dimensional range queries in sensor
networks, ACM SEYSYS 03
16
Q A