Improving Mobile Station Energy Efficiency in IEEE 802.16e WMAN by Burst Scheduling

1

Improving Mobile Improving Mobile Station Energy Station Energy

Efficiency in IEEE Efficiency in IEEE 802.16e WMAN by 802.16e WMAN by Burst SchedulingBurst Scheduling

Jinglin Shi, Gengfa Fang, Yi Sun, Jihua Zhou, Zhongcheng Li, and Eryk Dutkiewic

zIEEE GLOBECOM 2006

2

OutlineOutline• Introduction• System model• Longest virtual burst first scheduling• Simulation• Conclusion

3

IntroductionIntroduction• Wireless Network Interface (WNI)• MS has two states

– Awake mode and sleep mode

• Save power by turning off the WNI• Each sleep cycle is divided into

multiple sleep intervals• Each interval is made up of a sleep

window and a listening window

4

Energy saving factorsEnergy saving factors• IEEE 802.16e QoS requirements

– Minimum data rate• Inter-state transition takes extra time an

d energy

5

System modelSystem model• TDMA is used where bandwidth is calcul

ated in time slots• The uplink and downlink traffic is separa

ted in the TDD mode• We only consider downlink scenario fro

m BS to MSs• Data rate is fixed for all MSs• Minimum data rate as the only QoS requ

irement of each MS

6

NotationsNotations• M: the number of MSs included in one cell system• i: index of users in the cell, (i=1,2,…,M)• n: the index of time slot, (n=1,2,…)• rn

i: date rate that MS i allocated by time slot n• Rmin

i: minimum data rate which is the QoS of user i• sn

i: the state of MS i in time slot n, (1: awake; 0: sleep)• Paw: average energy consumed in each time slot by each

MS in awake state• Ptn: average energy consumed of state switch

7

Formulate energy conserving Formulate energy conserving scheduling problemscheduling problem

8

Energy consumption Energy consumption analysisanalysis

• When MS awake, scheduler must allocate as many time slots as possible

• Scheduler must choose relatively better channel quality MS for each time slot

• When MS is sleeping, it shouldn’t be awaked unless it will violate QoS requirement

9

Longest virtual burst first Longest virtual burst first scheduling (LVBF)scheduling (LVBF)

• Virtual burst: a period of time where there are one primary MS and multiple secondary MSs sharing the time slots

• Primary MS: chosen from awake MSs and occupies almost all the bandwidth during the period

• Secondary MS : all the other awake state MSs except the primary MS and they have enough resource for their minimum data rate

10

Example of virtual burst Example of virtual burst schedulingscheduling

MS i starts sleep request when rni > Rmax

i

11

DDefinitionefinition analysis analysis• Choose primary MS

• Idle Rateξ : rate of idle time slots to the total time slots for the primary MS (degree of primary MS occupies the bandwidth)

A set of secondary MSs

Total channel capacity

12

DDefinitionefinition analysis (cont.) analysis (cont.)• Ending each virtual burst when:

• Sleep duration

ε is a System parameter, which trades off average delay and energy efficiency.

The size of the sliding window

13

DDefinitionefinition analysis (cont.) analysis (cont.)• Invoke the sleep-state MS i when

• The scheduling result

(non-empty)

14

LVBD scheduling policyLVBD scheduling policy

15

Example of LVBD scheduling Example of LVBD scheduling - Step 1- Step 1

• Start a burst and choosing the primary MS and the remaining awake-state MSs are secondary MS

MSi Rmaxi rn

i Rmini

1 17 5 2

2 20 7 4

3 14 3 1

MSi Rmaxi - rn

i

1 17-5=12

2 20-7=13

3 14-3=11min

primary MSi = 3

secondary MSi= {1,2}

16

Example of LVBD scheduling Example of LVBD scheduling - Step 2- Step 2

• Scheduling for the current time slot among the primary and secondary MSs

MSi ri Rmini

1 5 2

2 7 4

(non-empty)

Swk = { i: 1 and 5 < 2 2 and 7 < 4 }

empty

Secondary MS 1

Secondary MS 2

Primary MS 3

(i* : Primary MS)

17

Example of LVBD scheduling Example of LVBD scheduling - Step 3- Step 3

• Update MSs’ perceived data rate base on the scheduling result in Step 2

MSi ri Rmini

1 1 2

2 3 4

(non-empty)

Swk = { i: 1 and 1 < 2 2 and 3 < 4 }

non-empty

MSi (rni – Rmin

i )/Rmini

1 (1-2)/2=-0.5

2 (3-4)/4=-0.25

min

Secondary MS 1

Secondary MS 2

Primary MS 3

(i* : Primary MS)

18

Example of LVBD scheduling Example of LVBD scheduling - Step 4- Step 4

• If current primary MS goes into sleep state, start a new virtual burst and go to Step 1, otherwise go to Step 5

MSi Rmaxi rn

i Rmini

3 14 15 2

rni > Rmax

i

15 > 14starts sleep request

Calculate the sleep durationIf Lsw = 3

(1- (dni* / 3) 15 = 2

dni* = 3 (1- 2/15) = 2.6

go to Step 1 and MS 3 sleeps for 2.6 time slot

19

Example of LVBD scheduling Example of LVBD scheduling - Step 5- Step 5

• If the event (ξ > ε) happens, start a new virtual burst and go to Step 1, otherwise, go to Step 2 for the next scheduling cycle

Calculate the idle rate

MSi ri Rmini

1 5 2

2 7 4

If C = 30, ε=0.5

ξ=(2+4)/30=0.2

ξ = 0.2 < 0.5 = ε

go to Step 2

20

Example of LVBD scheduling Example of LVBD scheduling (cont.)(cont.)

Secondary MS 1

Secondary MS 2

Primary MS 3rn

i > Rmaxi

Sleep MS 3

Primary MS 1

Secondary MS 3

ξ > ε

Primary MS 2

21

SimulationSimulation• Compare to Round Robin algorithm• Fading channel is represented by nine-

state Markov chain• Generated traffic in BS as a Poisson

process• Each packet is fixed size• Average energy efficiency as

• Each state transition cost 100 time slots unit of energy

22

Average energy efficiency vs syAverage energy efficiency vs system payloadstem payload

23

Minimum data rates Minimum data rates guarantee for usersguarantee for users

24

ConclusionConclusion• Proposed a Longest Virtual Burst First sc

heduling algorithm• LVBF prolongs MSs’ lifetime by reducin

g the average time when MSs stay in awake state and state transition times

25

Thank youThank you