Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 22

An Opportunistic Uplink Scheduling Scheme to

Achieve Bandwidth Fairness and Delay for


Multiclass Traffic in Wi-Max (IEEE 802.16)
Broadband Wireless Networks

Hemant Kumar Rath, Abhijeet Bhorkar, Vishal Sharma


Dept. of Electrical Engg., IIT-Bombay
{hemantr,bhorkar,vsharma@ee.iitb.ac.in}

IEEE Globecom – 2006


San Francisco, CA

©Copyright 2005-2006
All Rights Reserved
Motivation

 Request-grant mechanisms, service types defined


in std.
 Request is either in Contention mode or Contention free
(Polling) mode
 Service types need QoS in terms of delay guarantees

 Scheduling mechanisms are not defined


 Scheduling in both uplink and downlink is open
 Providers/vendors can have their own scheduling algos.

 Scheduling mechanism must balance....


 Fairness in bandwidth alloc. with delay guarantees

©Copyright 2005-2006 2
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Motivation (cont’d)

 Polling mode
 Poll each SS in every frame or in every k frames

 Polling interval k is a function of


 Delay tolerance Td
• UGS: 10ms, rtPS: 50ms, nrtPS: 200ms, BE: 500ms
 Fairness measure
 System efficiency

 Provider selects k to balance efficiency & fairness


 k may depend upon class of traffic

©Copyright 2005-2006 3
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Uplink Scheduling Scheme

R
eq
ue

st s
SS4 st SS1
s

ue
G

nt s
Req
ra
n

Gra
ts
Requests Requests

Grants Grants

SS3 SS2
BS

©Copyright 2005-2006 4
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Optimum Polling Interval k

 Polling mode
 BS polls each SS every k frames
 Worst case fairness is better if polled in every frame
 Normalized delay is better if polled in some k frames

 Design problem is to find an optimum k

 Approach: Minimize weighted sum of


 Normalized delay
 Worst case fairness in bandwidth allocation

©Copyright 2005-2006 5
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Opportunistic Scheduling

SINR1 q1(t)
SS1
SINR2 d1(t)

BS q2(t)
SS2
SINRk d2(t)
Scheduler

qk(t)
SSk
BS takes into account dk(t)
 Channel characteristics [ SINR1 , SINR2 , SINRk ]
 Queue lengths [q1 (t ), q2 (t ), qk (t )]
 Delay counters [d1 (t ), d 2 (t ), d k (t )] at scheduling instant, based on COS

©Copyright 2005-2006 6
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Opportunistic Deficit Round Robin
(O-DRR)

 Channel is static in a frame interval

 Slot assignment is opportunistic


 Assign slots only if channel is good and flow is active

 DRR variant for slot assignment


 Use queue state, delay requirements and lag/lead info.

 Works for single- and multi-class traffic


 SS with large Td relinquishes resources to SS with small Td

©Copyright 2005-2006 7
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
O-DRR Uplink Scheduling

SS6 SS1
SS2

SS5

Scheduling Scheduling Scheduling SS3


Instant Instant Instant
SS4

Tf

kTf
Scheduling Scheduling Scheduling
Epoch Epoch Epoch

©Copyright 2005-2006 8
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Scheduling Multiclass Traffic

 Number of slots assigned to an SS depends upon


 Delay counter
• How close a HOL packet is to its delay bound
• Weight is more if it closer to the delay limit

d i  Td ( i )  nT f
 Deficit counter
• Weight is more if the deficit counter is high

 Weights w
  1/delay counter
  deficit counter

©Copyright 2005-2006 9
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
O-DRR Uplink Scheduling

d1=10, d2=30, d3=25, d4=20 SS1 SS2


SS6
SS1=28, SS2=6, SS4=15, SS6=11

Schedule: weights (wi)


and lag/lead counter
Schedulable Set SS5

{SS1,SS2 ,SS4, SS6} SS3


SS4

Tf

kTf
Scheduling
Epoch
Eligible Set
{SS1, SS2, SS4, SS6}
SINRi > SINRth & Backlogged

©Copyright 2005-2006 10
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
O-DRR Uplink Scheduling

d1=10, d2=30, SS1 SS2


SS6
d3=25, d4=20
SS1=23, SS2=5,
d1=5, d2=25
SS4=13, SS6=9
SS1=46, SS2=14
Sch Set SS5
Sch Set
{SS1,SS2,
{SS1,SS2} SS3
SS4,SS6} SS4

Tf

kTf
Scheduling
Epoch
Eligible Set
{SS1, SS2, SS4, SS6}
SINRi > SINRth
& Backlogged
©Copyright 2005-2006 11
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
O-DRR Uplink Scheduling

SS6 SS1
SS2

Sch Set
{SS1,SS2, Sch Set Sch Set SS5

SS4, SS6} {SS1,SS2} {SS2,SS6} SS3


SS4

Tf

kTf
Scheduling
Epoch
Eligible Set
{SS1, SS2, SS4 , SS6}
SINRi > SINRth
& Backlogged
©Copyright 2005-2006 12
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
O-DRR Uplink Scheduling

SS1
SS6 SS2

SS5

SS3
SS4

Tf

kTf
Scheduling Scheduling
Epoch Epoch
Eligible Set Eligible Set
{SS1, SS2, SS4, SS6} {SS2, SS3, SS4, SS6}
SINRi > SINRth
& Backlogged
©Copyright 2005-2006 13
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Simulation Setup

 No. of users = 100


 No. of classes = 2
 k = 75, 100

 All flows backlogged (heavy traffic assumption)


 Delay requirements
 Class1 = 200ms
 Class2 = 500ms

 Total no. of frames scheduled = 2000


 Uplink slots per frame = 100

 Drop packets only if delay is violated

©Copyright 2005-2006 14
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Fairness and Throughput

 O-DRR is fair
 Fair among users
• Max. difference in allocated bandwidth < 10 % of average
 Fair among traffic classes
• Both class1 and class2 traffic get almost equal number of slots
 As k increases, fairness decreases (intuitively expected)

©Copyright 2005-2006 15
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Delay Performance

 Meets delay guarantees of different classes of traffic


 Packets are dropped only if delay is violated
 Packet drop is less than 8.5% for both classes of traffic
 For larger k, the dropping percentage is higher
• For worst case k=100, 91.5% of traffic meets its delay

©Copyright 2005-2006 16
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Choosing Polling Interval k

 Jain’s fairness index is


more than 95%
 A series of k are tested for
fairness

 Possible to trade off


fairness for delay

 Appropriate k to satisfy
• Fairness & bandwidth
requirements

©Copyright 2005-2006 17
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Discussion

 Low complexity scheduling algorithm

 The scheduling is done in the MAC layer


 It is a cross layer scheduling scheme involving PHY and
MAC layer

 Jain's fairness index remains above 90%


 It is possible to tradeoff fairness for delay

 O-DRR ensures delay requirements of users

©Copyright 2005-2006 18
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Future Work

 Multi-rate users (SSs) based on channel condition


 Adaptive to channel condition where SS can select a
particular modulation scheme and data rate

 Effect of location-dependent channel variations

 Stability analysis of the individual queues

©Copyright 2005-2006 19
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
©Copyright 2005-2006
All Rights Reserved
Example of O-DRR Scheme
 Assumptions
 Total no of slots = 60
 Number of users = 6
 Per user (quantum) = 10
 Tf= 5, K = 3

SS Cl SNR Qstate DRR Lag/Lead di Wi Slots Lag/Lead


Flag (before) assigned (after)

1 1 31 1 1 30 10 0.46 28 12

2 2 30 1 1 20 30 0.10 6 24
3 1 20 1 0 -35 20 0.0 0 -25

4 2 35 1 1 40 25 0.25 15 35

5 1 23 1 0 15 18 0.0 0 25

6 2 32 1 1 23 20 0.18 11 22

Scheduling Epoch1, Scheduling Instant1


©Copyright 2005-2006 21
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved
Example of O-DRR Scheme

SS Cl SNR Qstate DRR Lag/Lead di wi Slots Lag/Lead


Flag (before) assigned (after)
12
1 1 31 32 1 1 1 10 5 0.77 46 -24

24
2 2 30 34 1 1 1 10 25 0.23 14 20
-25
3 1 20 22 1 0 0 10 15 0 0 -15
35
4 2 35 25 1 1 0 10 20 0 0 45
25
5 1 23 24 1 0 0 10 13 0 0 35
22
6 2 32 21 1 1 0 20 15 0 0 32

Scheduling Epoch1, Scheduling Instant2

©Copyright 2005-2006 22
IEEE Globecom-2006, NXG-02: Broadband Access
All Rights Reserved

You might also like