1. Sriram Lakshmanan Zhenyun Zhuang
A Unified Architecture for the Design and Evaluation of Wireless Fair Queueing Algorithms
Sriram Lakshmanan
Zhenyun Zhuang

2. Introduction
In wireline networks resource reservation and allocation are typically used for QOS
Difficulty of performing reservation and fair scheduling in wireless networks due to
Location-dependent and bursty errors
Channel contention
Joint scheduling of uplink/downlink flows

3. Motivation for a unified Framework
Need for unified Architecture
Over the recent past, several algorithms have been proposed for wireless scheduling
Notion of fairness for the wireless domain unclear
Identifies Generic blocks
Also enables comparison of algorithms

4. Wireless Channel Model
Packet cellular network
Single-hop network
Neighboring cells transmit on different logical channels
Every mobile host in a cell can communicate with the base station
but two hosts are not necessarily within their ranges
Each stations performs the scheduling of both uplink and downlink transmissions in its cell

5. Key Characteristics of Wireless Channel
Wireless channel capacity is dynamically changing
Channel errors
Contention for channel among hosts
Hosts don’t have global knowledge
Scheduling has to take care of uplink and downlink
processing power and battery constraints

6. Service Model- Fluid Fair Queueing
Properties
Fairness among backlogged flows over infinitesimal time
Bounded delay channel access
Guaranteed minimum throughput for backlogged flows
In Summary: Full separation between flows
Minimum guarantees provided for a flow is unaffected by others
Assumes: Error free and no location dependent
All flows can transmit at a given time
Or none can transmit

7. Fairness criterion Each flow i, j satisfies the following property:
 i, jB(t1,t2)
Wi(t1, t2)
Wj(t1,t2) ri rj
Eq.(1)
B(t1, t2): set oh backlogged flows
Wi(t1, t2): channel capacity granted to each flow i
[t1, t2]: time interval

8. Wireline fair queuing fails over wireless channels (Illustration 1)
Flows f1, f2, f3 in [0,2] with r1=r2=r3
Error –free channel W3=1/3
Error –free channel W2=1/3
Error –free channel W1=1/3
[1, 2]
Error –free channel W2=5/6
Error –free channel W1=5/6
[0, 2]
Channel error W3=0
Error–free channel W2=1/2
Error–free channel W1=1/2
[0, 1) f3 f2 f1
flow
Time
Does not satisfy the fairness property of Eq.(1)

9. Wireline fair queuing fails over wireless channels (illustration 2)
Flows f1, f2, f3 in [0,2] with r1=r2=r3
Error –free channel W3=2/3
Error –free channel W2=0
Error –free channel W1=1/3
[1, 2]
Error –free channel W2=2/3
Error –free channel W1=2/3
[0, 2]
Channel error W3=0
Error–free channel W2=2/3
Error–free channel W1=1/3
[0, 1) f3 f2 f1
flow
Time
- Is it acceptable for f1 to be impacted?
- Over what time period should f3 be compensated for its loss?
- In this scheme separation of flows is lost.
- Tradeoff between separation and compensation

10. Wireless Fair Service Model
Short-term fairness
for backlogged flows when perceive a clean channel
Long-term fairness
for backlogged flows with bounded channel error
Channel-conditioned delay bounds for packets
Short-term throughput bounds
for flows with clean channels

11. Wireless Fair Service Model - Continued
Long-term throughput bounds
for all flows with bounded channel error
Support for both delay sensitive and error sensitive
Very useful for handling both delay sensitive and error sensitive flows in error-prone channels
Optionally, optimization of the schedulable region by decoupling the delay and bandwidth requirements of flows

12. Definitions in wireless fair queueing algorithms
Error-free service: service it would have received if all channels had been error-free
Leading flow: received channel allocation excess of it’s error-free service
Lagging flow: received channel allocation less than it’s error-free service
In sync flow:its channel allocations exactly the same as its error-free service

13. Unified Wireless Fair Queueing Framework
Emulate fluid fair queueing when all flows perceive error-free channels
Its goal is to minimize:
Wi(t1, t2)
Wj(t1,t2) ri rj

14. Seven Algorithms CSDPS IWFQ WPS
Channel State Dependent Packet Scheduling
IWFQ
Idealized Wireless Fair Queueing
WPS
Wireless Packet Scheduling
CBQ-CSDPS: Enhanced Class Based Queueing
SBFA
Server Based Fairness Approach
WFS
Wireless Fair Service Algorithm
CIFQ
Channel-condition Independent Fair Queueing

15. Wireless fair queuing algorithms
The different proposals differ in terms of :
how the channel access swapping takes place
Between which flows the swapping tacks place
How the compensation model is structured

16. Five Components Error-free service Lead and lag model
Compensation model
Slot queues and packet queues
Channel monitoring and prediction

18. 1. Error-free service How much service a flow should receive
In an ideal error-free channel environment
Packetized approximation of fluid fair queueing
WFQ by IWFQ, SBFA
WRR by WPS …..

19. WFQ

20. 2. Lead and Lag Model Lag of a lagging flow Lead of a leading flow
The amount of additional service to which it is entitled in the future in order to compensate
for lost service in the past
Lead of a leading flow
The amount of additional service that the flow has to relinguish in the future
In order to compensate for additional service in the past
Two approaches to compute
Difference between the error-free service and received
Number of slots allocated to the flow during its channel error
Upper bounded

21. 3. Compensation model Determines: Three main issues has to address:
how lagging flows make up their lag
how leading flows give up their lead
Three main issues has to address:
When does a leading flow relinquish the slots that are allocated to it ?
All slots till it becomes in sync.(IWFQ)
Fraction of slots.(constant CIF-Q proportion WFS)
Never, assume there is some reserved BW.(SBFA)

22. Compensation model (cont)
When are slots allocated for compensating lagging flows ?
Preferentially allocated till there is no lagging flow. (IWFQ)
Allocated only when leading flows relinquish slots. (CIF-Q & WFS)
Allocated from reserved BW. (SBFA)
How are compensation slots allocated among lagging flows ?
Flow with the largest lag. (CIF-Q)
According to the order they became backlogged. (IWFQ & SBFA)
In proportion to its lag. (WFS)

23. 4. Slot queues and packet queues
Allow for the support of both delay sensitive and error sensitive flows in a single framework
Decouple connection-level packet management policies from link-level packet scheduling policies
Each flow has 1 slot queue and 1 packet queue

24. Cont’ Slot queues: Packet queues:
Slot is the unit of channel allocation
Delay-sensitive flows
Delete packets once it violates its delay bound
Packet queues:
Packet is the unit of data transmission
Error-sensitive flows
Do not delete head-of-line packet upon channel error during transmission

25. 5. Channel monitoring and Prediction
Perfect channel-dependent scheduling depend on accurate channel state
Errors typically occur over bursts and highly correlated in successive slots
Channel prediction can be achieved by using:
N-state Markov model
One step prediction algorithm

26. One Example:IWFQ Idealized Wireless Fair Queueing
Error-free
WFQ
Lag and Lead
The difference between service tags
Compensation Model
Favors lagging flows
But may starve leading flows in the short run
Comments
Short-term Fairness and Throughput Bounds are violated
Provides Long-term fairness and bounded delay channel access

27. Simulations WFS and CIF-Q achieve all the properties
Short-term and long-term fairness
Short-term and long-term throughput bounds
Tight delay bounds for channel access