QoS Control For Sensor Networks Iyer, R.; Kleinrock, L. ICC 2003.

QoS Control For Sensor Networks

Iyer, R.; Kleinrock, L.Iyer, R.; Kleinrock, L.

ICC 2003ICC 2003

outlineoutline

►IntroductionIntroduction►Problem descriptionProblem description►The GUR GAMEThe GUR GAME►Network model and algorithmNetwork model and algorithm►SimulationSimulation►ConclusionConclusion

IntroductionIntroduction

►Sensor networks are made up of small devices Sensor networks are made up of small devices equipped withequipped with ProcessorsProcessors MemoryMemory Short-range wireless communicationShort-range wireless communication

►Sensor network have severe energy constraintsSensor network have severe energy constraints Preservation of battery powerPreservation of battery power

IntroductionIntroduction

►What is sensor network QoS?What is sensor network QoS? Sensor network resolutionSensor network resolution An An optimaloptimal number of sensors sending information number of sensors sending information

toward information-collecting sinks, typically base toward information-collecting sinks, typically base stationsstations

Problem descriptionProblem description

►We want to accomplish two thingsWe want to accomplish two things MaximizeMaximize the the lifetimelifetime of the sensor network of the sensor network

►Sensors periodically power-down to conserve their battery Sensors periodically power-down to conserve their battery energyenergy

Have Have enoughenough sensors power-upsensors power-up and sending packets and sending packets toward the information sinks so that enough data is toward the information sinks so that enough data is being collectedbeing collected

Problem descriptionProblem description

►One base station with a broadcast channel to all One base station with a broadcast channel to all the sensors the sensors

►It knows the It knows the optimal numberoptimal number of sensors that of sensors that should be power-on and sending packets at any should be power-on and sending packets at any given timegiven time

►Broadcast the probability Broadcast the probability p*(t)=(optimal number / total number) at time t to all p*(t)=(optimal number / total number) at time t to all

sensorssensors

The GUR GAMEThe GUR GAME

►We have many players and a refereeWe have many players and a referee None of them are aware of the othersNone of them are aware of the others

►Every second, the referee asks each player to vote yes or no, then counts up the yes and no answers

►A reward probability r = r(k) is generated as a function of the number k of players who voted yes

►0 ≤ r(k) ≤ 1

The GUR GAMEThe GUR GAME

►The reward function r = r(kt) =0.2 + 0.8 ev

v = -0.002 (kt – 35)2

kt = the number of yes received at time t

The GUR GAMEThe GUR GAME

►Each player, regardless of how he or she voted, is then independently rewarded (with probability r) penalized (with probability 1-r).

►Assume that at some point the number of players voting yes was k1. The reward probability would be r(k1)

The GUR GAMEThe GUR GAME

The GUR GAMEThe GUR GAME

►Note that the maximum of the example occurs at k* = 35

►No matter how many players there are, we can “construct” them in such a way that approximately k* of them

The GUR GAMEThe GUR GAME

►Each player has a memory of his previous trials►we associate with each player j, a finite discrete-

time automaton Mj►The player is allowed to be in only one state at

any given time.

The GUR GAMEThe GUR GAME

The GUR GAMEThe GUR GAME

►The player votes Yes when he is in a positive numbered state No when he is in a negative numbered state.

►When in a negative (positive) numbered state Leftward (rightward ) if he is rewarded Rightward (leftward) when he is punished

Network model and algorithmNetwork model and algorithm

► We associate with the base station a Gur reward function r(k)

► At each time t, the base station counts the number of packets kt it has received from the sensors

► It then calculates the Gur reward probability r(kt )► Finally, it broadcasts this probability to all the sensors.► Each sensor, in turn, independently rewards itself with

probability r(kt ) and punishes itself with probability 1-r(kt )

Network model and algorithmNetwork model and algorithm

►assume that we have a collection of m sensors S1 through Sm and one base station B

►The sensor Si will power-up when it is in a positive numbered state power-down when it is in a negative numbered state

►if a sensor is powered-up, it will send a data packet containing sensor information toward the base station

SimulationSimulation

►Memory size N is equal to 1►100 sensors in the network►Each sensor picks a random state as its initial

state►base station desires a rate of 35 packets received

at each time t

SimulationSimulation

SimulationSimulation

SimulationSimulation

ConclusionConclusion

►quality of service (QoS) for sensor networks remains largely open, and it is a non-trivial problem to specify the optimum number of sensors

►In this paper we presented an algorithm using the Gur Game paradigm that allowed the base station to specify the optimal number of sensors

Thank youThank you