Fair Real-time Traffic Scheduling over Wireless Local Area Networks

Insik ShinJoint work with M. Adamou, S. Khanna, I. Lee, and S. Zhou

Dept. of Computer & Information Science

University of Pennsylvania

Real-Time Packet Scheduling

• Real-Time Flow– Periodic interval

• interval between arrival time of two packets– Deadline

• a packet should be scheduled and successfully transmitted within the time, otherwise it is lost

– Acceptable packet loss rate• degradation = actual loss rate – acceptable loss rate

Scheduling over Wireless LAN

• Cellular Wireless Network – one base station (BS)

– multiple mobile hosts (MHs)

– BS schedules real-time packet

transmissions of BS & MHs

using polling mechanism

BSMH

MHMH

Scheduling over Wireless LAN

• Cellular Wireless Network

– Unpredictable

channel error •

location dependent

• burstyBS

MH1

MH3MH2

Scheduling Motivation

• Unpredictable wireless channel error– failure of packet delivery in time– degraded quality of service– some flows may have more degraded QoS

while others may have less degraded QoS, due to location dependent property

• Fair scheduling of real-time packets with deadlines in the presence of the errors

Previous Work

• QoS guarantees over wireless links– No consideration of fairness issue

• WFQ over wireless networks– No consideration of deadline constraint

• (m,k)-firm deadline model– should meet deadlines of m out of k consecutive

packets

– Similar to our deadline model, except that we consider fair degradation without any guarantees in wireless network (unpredictable error can violate any guarantee)

Scheduling Objectives

1. Achieving fairness by minimizing the maximum degradation among all flows

2. maximizing the overall system throughput simultaneously

• Online scheduling algorithm– without knowledge of error in advance

Theoretical Results

• No online optimal algorithm for our scheduling objectives– for throughput maximization, an online algorithm can

achieve a performance ratio of two w.r.t. the optimal

– for achieving fairness, no online algorithm can guarantee a bounded performance ratio w.r.t. optimal

– Hence, none can guarantee a bounded performance ratio w.r.t. optimal for the combined objectives

• A polynomial time offline algorithm that optimally achieves our scheduling objectives

Online Scheduling Algorithms

• EDF (Earliest Deadline First)– Naturally suited for maximizing throughput

• GDF (Greatest Degradation First)– Seeks to minimizing the maximum degradation

• EOG (EDF or GDF)– Simply combines EDF and GDF

• LFF (Lagging Flows First)– Favors lagging flows (receiving degraded QoS)

in a more clever, sophisticated manner

Online Algorithm - LFF

• LFF (Lagging Flows First)– Try to schedule the k most lagging flows when

at most k flows can be scheduled in the next available slots.

Scheduling Example

available for schedule NOT available for schedule

slot 1 2 3 4

schedule ? ? ? ? Flow i

1 0.1

2 0.2

3 0.6

4 0.7

5 0.8

6 0.9

slot1 2 3 4

Assume that i decreases by

upon a successful transmission of a packet of flow i and increases by upon a failure of a packet, where 0.05 < < 0.1

i – degradation degree of flow i

Scheduling Example

scheduled

Flow i

1 0.1

2 0.2

3 0.6

4 0.7

5 0.8

6 0.9

slot1 2 3 4

slot 1 2 3 4

schedule 2 1 4 6

EDF schedulei

max:

schedule 4 packets

Scheduling Example

scheduled

Flow i

1 0.1

2 0.2

3 0.6

4 0.7

5 0.8

6 0.9

slot1 2 3 4

slot 1 2 3 4

schedule 6 5 4

GDF schedulei

GDF – Greatest Degradation First

max:

schedule 3 packets

Scheduling Example

scheduled

Flow i

1 0.1

2 0.2

3 0.6

4 0.7

5 0.8

6 0.9

slot1 2 3 4

slot 1 2 3 4

schedule 2 6 4 5

EOG schedulei

EOG– EOF or GDF

max:

schedule 4 packets

Scheduling Example

scheduled

Flow i

1 0.1

2 0.2

3 0.6

4 0.7

5 0.8

6 0.9

slot1 2 3 4

slot 1 2 3 4

schedule 3 4 5 6

LFF schedulei

max:

schedule 4 packets

Error Handling Mechanisms

• Re-scheduling Mechanisms1. No re-scheduling - dropping packets with errors

2. Immediate re-scheduling - ignoring errors

3. Delayed re-scheduling – How long does it need to delay?– Backoff value = deadline/2

Simulation

• Performance Metrics

1. Degradation (for each flow)– Fraction of packets lost beyond the acceptable

packet loss rate

2. Throughput (over all flows)– Fraction of successfully transmitted packets

Results – Max Degradation

0

0.1

0.2

0.3

0 0.1 0.2 0.3 0.4Error Duration Rate

Deg

rada

tion

deg

ree

EDFGDFEOGLFF

Results – Throughput Ratio

0.98

0.985

0.99

0.995

1

1.005

1.01

1.015

1.02

0 0.1 0.2 0.3 0.4Error Duration Rate

Thr

ough

put r

atio

vs E

OG

EDFGDFEOGLFF

Conclusion• Our scheduling objectives

1. Fairness – minimizing the maximum degradation

2. Overall throughput maximization• Our theoretical study showed that no online

algorithm can be guaranteed to achieve a bounded performance ratio for fairness objective

• For fairness objective1. LFF 2. GDF 3. EOG 4.EDF

• For maximum throughput objective1. EDF 2. LFF 3. EOG 4.GDF