DataCentric Storage in Sensornets with GHT, a Geographic Hash Table

1
Data-Centric Storage in Sensornets with GHT, a
Geographic Hash Table

• Scott .Shenker, Deborah Estrin et al.
• Appeared in Mobile Networks and Application,2003

Presented by Chih-chieh Hung 2004.8.3
2
Outline

• Introduction
• Sensornet Architecture
• Data-centric Storage
• Conclusion

3
Introduction

• In sensornet, the content of data is more
  important than the identity of the nodes that
  gather them.
• It is important to develop an efficient method to
  allocate the data in sensornet.

4
Sensornet

• Assumptions
• Sensornet Data
• Data-dissemination Methods

5
Assumptions

• Nodes know their geographic location.
• The sensornet is connected to the outside world
  through the sensornet to the access point.
• Energy is limited for sensornet nodes.

6
Sensornet Data

• Type of Information
• Observations
• The low-level readings from these sensors
• Events
• pre-defined constellations of low-level
  observations

7
Sensornet Data (contd)

• Operations used to manipulate data
• Tasks
• User send instruction to nodes to run
  certain local identification tasks.
• ex taking temperature readings.
• Actions
• External store
• Local store
• Data-centric store
• Queries
• Used to elicit the event information from
  the sensornet

8
Data-dissemination Methods

• External Storage (ES)
• Local Storage (LS)
• Data-centric Storage (DCS)

9
External Storage (ES)

• Upon detection of events, the relevant data are
  sent to external storage where they are further
  processed as needed.
• No cost for external queries.
• Approximate communication cost
• (total number of event detected) n1/2

10
Local Storage (LS)

• Event information is stored locally (at the
  detecting node) upon detection of an event.
• Queries are flooded to all node at a cost of
  O(n).
• Communication Cost Qn Dq n1/2
• Q the number of queries in total.
• Dq the number of events detected for each event
  
• queried for.

11
Data-centric Storage (DCS)

• After an event is detected the data are stored by
  name within the sensornet.
• Query Cost O(n1/2)
• Communication Cost O(n1/2)

12
Data-centric Storage

• DCS problems
• Storage abstraction
• Design criteria for scalable, robust DCS
• Geographic hashing table
• Algorithm
• GPSR
• Home node and home perimeter
• Perimeter refresh protocol
• Structured replication

13
Storage Abstraction

• DCS provides a (key, value)-based associative
  memory.
• 2 operations DCS supports are
• Put(k, v)
• stores v (the observed data) according to the
  key k, the name of data
• Get(k)
• retrieves whatever value is stored associated
  with the key k

14
Design Criteria for Scalable, Robust DCS

• Sensornets represent a particularly challenging
  enviroment for a distributed storage system
• Node failures
• Topology changes
• System scale in nodes
• Energy constraints

15
Design Criteria for Scalable, Robust DCS (contd)

• Design Criteria for Ensuring scalability and
  robustness
• Persistence
• Consistency
• Scaling in database size
• Scaling in node count
• Topological generality

16
Geographic Hashing Table (GHT)

• The core step in GHT is the hashing of a key k
  into geographic coordinations.
• GHT uses perimeter refresh protocol to provide
  both persistence and consistency.
• GHT avoids creating a hotspot of communication
  and storage by structured replication.

17
GPSR

• Under GPSR, packets are routed geographically.
• Two distinct algorithms for routing
• Greedy forwarding algorithm
• Perimeter forwarding algorithm

18
Greedy Forwarding Algorithm

• Greedy forwarding rule

x
D
y
19
Greedy Forwarding Algorithm (contd)

• Greedy forwarding algorithm maybe failed

D
z
y
y
w
x
20
Perimeter Forwarding Algorithm

• Applying right-hand rule.
• Ex
• GPSR routes perimeter mode packets on a planar
  subgraph of network connected graph.

2
x
z
3
1
y
21
State Transition Diagram of Two Forwarding
Algorithms
Greedy forwarding fails
Perimeter Mode
Greedy Mode
Packet reaches a node closer to destination
reach destination
reach destination
Stop
22
Home Node and Home Perimeter

• Home node
• the node geographically nearest the destination
  coordinates of the packet
• Home perimeter
• The packet traverses the home perimeter that
  encloses the destination.
• Under GHT, the home node knows to consume the
  packet when it returns.

23
Perimeter Refresh Protocol

• Perimeter refresh protocol (PRP) stores a copy of
  a key-value pair at each node on the home
  perimeter.
• Replica nodes
• Nodes on the home perimeter except the home
  node.

24
Mechanism of PRP Refresh Packet

• PRP uses refresh packets to ensure the node
  closest to a keys hash function will be the home
  node.

other
home
25
Mechanism of PRP Refresh Packet (contd)
New home node
originator
26
Providing Persistence Mechanism
Death Timer Td
Cache Reser Tt ?Td
Not originator
Not originator
Tt 0 Home?
GHT routing
Replica Node
27
Providing Persistence Mechanism
Death Timer Td
Cache Reser Tt ?Td
Not originator
Not originator
Tt 0 Home?
GHT routing
Replica Node
28
Structured Replication

• Use a hierarchical decomposition of the key space
• Storage cost ?
• Query cost ?

29
Conclusion

• GHT based on GPSR is a geographical routing
  scheme.
• GHT hashes keys into geographic coordinates, and
  stores a key-value pair at the sensor node
  nearest the hash of its key.
• GHT in DCS system ensures scalability, robustness
  and fault-tolerant.