rakocevic

Upload: deepakrajshetty

Post on 06-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/2/2019 Rakocevic

    1/12

    Sep 2004 QofIS 04 1

    QoS Constraints in Bluetooth-based Wireless Sensor Networks

    Veselin Rakocevic, Muttukrishnan Rajarajan,Kerry-Ann McCalla, Charbel Boumitri

    City University, London

  • 8/2/2019 Rakocevic

    2/12

    Sep 2004 QofIS 04 2

    Motivation

    QoS in sensor networks

    a change to the way we look at QoS

    Need for satisfying human users much smaller

    Independent, low-power nodes communicating

    They require:

    Connectivity

    Reliability (Connection robustness)

    Low-power operation FRESHNESS OF INFORMATION

  • 8/2/2019 Rakocevic

    3/12

    Sep 2004 QofIS 04 3

    Bluetooth

  • 8/2/2019 Rakocevic

    4/12

    Sep 2004 QofIS 04 4

    QoS Constraints in Bluetoothnetworks

    Interference

    Error Correction

    Transport Control

    Service Discovery, Connection Establishment,and Scatternet formation

    Power Consumption

    Scheduling

  • 8/2/2019 Rakocevic

    5/12

    Sep 2004 QofIS 04 5

    Scheduling

    Master is responsible for scheduling

    Bluetooth specification does not specify ascheduling scheme round-robin scheduling isassumed

    Lot of research being done on using differentscheduling paradigms

    This paper presents a comparison betweenseveral scheduling schemes in sensor networkenvironment

  • 8/2/2019 Rakocevic

    6/12

    Sep 2004 QofIS 04 6

    Simulation Model

    7 slaves and one master

    Slaves producing traffic Uniform file size we assume each slave produces a fixed

    amount of fixed-size packets (bursts)

    Master applying the scheduling schemes

    Three schemes: Exhaustive: master serves a slave until there are any

    packets in slaves buffer

    Limited Exhaustive: master serves a slave until there are

    any packets belonging to a single burst in the slavesbuffer

    Maximum Burst Delay First: master serves a slave that hasa burst which had waited for the longest time

    id id

  • 8/2/2019 Rakocevic

    7/12

    Sep 2004 QofIS 04 7

    ==

    7

    1

    11

    ii

    ii

    dN

    df

    ==

    7

    1

    11

    ii

    ii

    dN

    df

    Simulation Model

    Symmetric and asymmetric load

    Measurements:

    Percentage of flushed bursts

    Average burst delay Fairness index (for asymmetric load)

    =

    =7

    1

    11

    i

    i

    ii

    d

    N

    df

  • 8/2/2019 Rakocevic

    8/12

    Sep 2004 QofIS 04 8

    Results 1 symmetric load

    2

    3

    4

    5

    6

    7

    8

    9

    10

    1112

    1.00E-02 1.00E-03 1.00E-04 1.00E-05

    error probability

    percen

    tage

    offlushe

    dd

    ata

    bursts

    exhaustive

    BURSTMinimal burst delay

    In terms of flushprobability, MBDF

    is the best

    In terms of average

    delay, exhaustiveis the best

  • 8/2/2019 Rakocevic

    9/12

    Sep 2004 QofIS 04 9

    Results 2 asymmetric load

    -0.08

    -0.06

    -0.04

    -0.02

    0

    0.02

    0.04

    0.06

    0.08

    0.1

    1.E-02 1.E-03 1.E-04 1.E-05

    bit error probability

    delay

    fairnessi

    ndex

    MBDF LimExh ExhMBDF is better for thehigh-load node

    But how much?

    More results neededand also implementation

    details need to besolved

  • 8/2/2019 Rakocevic

    10/12

    Sep 2004 QofIS 04 10

    Results 3 asymmetric load

    -0.3

    -0.25

    -0.2

    -0.15

    -0.1

    -0.05

    0

    0.05

    0.1

    0.15

    1.00E-02 1.00E-03 1.00E-04 1.00E-05

    bit error probability

    flushp

    robability

    fairness

    index

    MBDF LimExh Exh

    Interesting exhaustive schedulinggives more protectionto the asymmetric node

  • 8/2/2019 Rakocevic

    11/12

    Sep 2004 QofIS 04 11

    Conclusions?

    QoS requirements in pervasive/ad-hoc/sensorsystems not the same as QoS requirementsfor web and voice applications

    Polling scheduling schemes very interesting toanalyse

    Our preliminary results do not agree withexpectations

    we need to work more

  • 8/2/2019 Rakocevic

    12/12

    Sep 2004 QofIS 04 12

    Low power service discovery andscatternet formation

    Using Bluetooth in a smart home environment

    Connectivity, connection robustness, and lowpower operation essential

    We are working on a Low Power Discovery

    Scheme Objectives: to guarantee connectivity to all nodes

    Nodes are in low-power SNIFF mode

    If a master disappears, nodes have to find eachother and establish new piconets

    Difficult to do if they are in low power mode, sincethey have to continuously listen to service discoverymessages