CORD : Energyefficient Reliable Bulk Data Dissemination in Sensor Networks - PowerPoint PPT Presentation

1 / 43
About This Presentation
Title:

CORD : Energyefficient Reliable Bulk Data Dissemination in Sensor Networks

Description:

Deliver object to subset of nodes. Deliver object to ... Cheng's Single leader Algo. S. Algorithm Extensions. Link Quality. ... Difference from Cheng's Algo ... – PowerPoint PPT presentation

Number of Views:71
Avg rating:3.0/5.0
Slides: 44
Provided by: jatinde
Category:

less

Transcript and Presenter's Notes

Title: CORD : Energyefficient Reliable Bulk Data Dissemination in Sensor Networks


1
CORD Energy-efficient Reliable Bulk Data
Dissemination in Sensor Networks
  • Leijun Huang, Sanjeev Setia
  • George Mason Univeristy
  • Presented By Jay Walia

2
Abstract
  • Deluge Reduce Latency
  • CORD Primary goal is to reduce energy
    consumption
  • 2 Phase approach
  • Deliver object to subset of nodes
  • Deliver object to remaining nodes
  • Installs a coordinated sleep schedule
  • Comparable latency compared to Deluge but
    significant energy conservation.

3
Bulk Data Dissemination
  • Over the air network programming is required
    after initial deployment
  • Reliable Bulk data dissemination protocol is
    required.
  • Deluge is the de facto standard

4
Categories of Protocols
  • 2 Categories
  • Propagated from sink to the rest of the network
    in a neighborhood by neighborhood fashion.
  • Deluge
  • Phased propagation
  • Propagate to Core Nodes
  • Propogate to remaining nodes
  • CORD

5
CORD Go Green
  • Energy conservation
  • Reduce control messages by building the core set.
  • Reduction in no. of competing nodes that transfer
    the object results in reduction in no. of message
    collisions.
  • Good for heterogeneous networks where subset of
    nodes is more powerful.
  • Aggressively use sleep schedule.

6
CORD Authors Contrib
  • Sleep scheduling 2 phase approach
  • Evaluation vs Deluge
  • First comprehensive evaluation

7
Deluge and others
  • MOAP, Deluge, MNP
  • Network reprogramming in Multi Hop networks
  • 3 phase handshaking
  • Selective NACKs and hop by hop error recovery
  • Objects into pages that allows pipelined page
    delivery

8
Sprinkler
  • 2 Phase
  • Connected Dominating Set
  • Packet level pipelining in the 1st phase
  • TDMA
  • No sleep scheduling
  • Core nodes forward every newly received data
    packets and piggybacks the negative acks and
    parent id, hence parent core nodes cannot sleep
    when child core nodes are tx.

9
CORD and GARUDA
  • Borrows from GARUDA
  • Garuda not optimized for large objects
  • Does not have pipelining.

10
CDS Connecting Dominating Set
  • Many algorithms exist.
  • CORD uses Chengs single leader algorithm
  • Virtual backbone construction in multi hop ad hoc
    wireless networks.

11
Overview
  • Core Construction Step
  • Either a core or a neighbor of core
  • 1st phase, transfer from sink to core, non core
    passively participate by listening to cores.
  • At end of 1st phase, non core already have
    significant portions of object
  • Second phase, non core asks for missing portions.

12
Design Tradeoffs
  • Sensor Networks are stationary and link quality
    of nodes is relatively stable.
  • Use the above information to reduce the no. of
    control messages while transferring the objects
    as 3 way handshake is not required.
  • Sleep scheduling most effective.
  • Conserve energy while still comparable latency.

13
Design
  • Core construction
  • Set up for each object dissemination, an
    infrequent event.
  • Chengs algorithm
  • Start with Non active White nodes
  • Dominators (core) Black Nodes
  • Dominatees (non core) Grey Nodes
  • Sink becomes dominator
  • White node marks itself grey and becomes
    dominated
  • Non active white nodes become active and contend
    for dominator
  • Maximum effective degree (no. of white neighbors)
    wins the competition and its grey parent also
    becomes dominator to complete CDS.
  • Next slide

14
Chengs Single leader Algo

S
15
Algorithm Extensions
  • Link Quality.
  • Use packet loss or LQI on CC2420
  • Threshold Qth
  • Quv min( Qu-gtv, Qv-gtu )
  • Coordinated Schedule at each node
  • Fixed slots - Repeated
  • Either a parent, child or quiescent (to become
    quiet).
  • Schedule is installed at time of core
    construction.

16
Schedule Installation
  • Sink initiates core construction
  • Sink sends CLAIM message announcing itself as a
    core node.
  • Time offset from the beginning of the first time
    slot to when the message was sent.
  • List of neighbors to whom sink has good links.

17
Schedule Installation
  • At Hop One
  • Nodes update their effective degrees by counting
    the neighbor not included in the sinks neighbor
    list.
  • If the node has a good link with the sink, it
    selects the sink as its parent in the core and
    initiates its own repeating schedule such that
    the start time of its own scheduled coincides
    with the slots of the sink.

18
Schedule Installation
  • Nodes that have sink as the parent broadcast
    their effective degree in the COMPETE message.
  • Nodes that hear the COMPETE message respond with
    a SUBSCRIBE message to the one with the maximum
    effective degree.
  • Node that receive the SUBSCRIBE message becomes
    the CORE node.
  • CORE nodes at level one that broadcast CLAIM
    messages.

19
Difference from Chengs Algo
  • In CDS, core nodes are selected sequentially
    where as in Chengs algorithm, they are selected
    at odd levels followed by at even levels.

20
Coordinated Node Sleep Scheduling

21
Coordinated Node Sleep Scheduling
22
Coordinated Node Sleep Scheduling
  • Longer time slots
  • Schedule coordination in CORD is more robust to
    clock drifts
  • Overhead of switching radio on/off is reduced.

23
2 Phase Data Dissemination
  • Core construction is followed by 2 phase step.
  • 1st phase
  • Pages are propagated in a pipelined fashion.
  • core node that has received one or more pages of
    the object broadcasts an advertisement with this
    information.
  • Its child nodes send the request for desired
    packets.
  • Parent broadcasts requested data packets.

24
2 Phase Data Dissemination
  • Requests are suppressed if another request is
    being sent for the same or lower numbered pages.
  • In 2nd Phase, non core nodes make requests to
    their local core nodes.

25
Difference from Deluge
  • Only core nodes are allowed to advertise and
    become data senders.
  • Nodes stick to fixed schedule
  • P, C and Q slots.

26
Protocol Evaulation
  • nesC - TinyOS
  • Measuring energy
  • Indirect method
  • Log no. of packet Tx and Rp, Radio on /off,
    EEPROM read/writes
  • Log post processed for consumption.
  • Used specifications in following table.

27
Energy Expenditure Table
28
Test Bed - Indoor
  • 2 different test beds
  • Indoor
  • 20 TelosB
  • 32 hour period, multiple experiments
  • No human/wireless LAN activity and lot of
    human/wireless LAN activity.
  • Node used to initiate dissemination of an object
    of the same size.

29
Indoor Test Bed
30
Test Bed - Outdoor
  • Outdoor
  • 33 Telos B
  • 3 11 grid
  • 2 meters row, 2.6 meters column

31
Outdoor Test Bed
32
Reception Rate
33
Object delivery Latency
34
Object size
  • 960 packets
  • 20 pages ( K 48 )
  • Node O as the sink.
  • 23 byte payload packet
  • Reception rate tested by HELLO messages after
    each run.

35
Empirical results
  • Object delivery latency independent of time of
    day ( b/n 200 270 )
  • 4 6 core nodes.

36
Results - Indoor
37
Results
  • 9 to 11 cores in outside test bed
  • 4-5 hops

38
Results - Outdoor
39
Individual Obj Dlv Latency
40
Uptime
41
Avg Pckt Tx
42
Simulation
  • Tossim and Power Tossim
  • Rectangular grid, random topology
  • Please refer paper for results
  • Or the dissertation

43
Conclusions
  • Energy consumption is 30 to 60 less compared to
    Deluge
  • Latency is comparable to Deluge
  • Nodes receive object earlier in CORD than Deluge.
Write a Comment
User Comments (0)
About PowerShow.com