QoS in ad hoc nets: distributed fair scheduling

SCOPE: Self-coordinating Localized FQ H. Luo et al “A Self-Coordinating Approach to Distributed FairQueueing in Ad Hoc Wireless

Networks”, Infocom 01

• Support mission critical applications– ad-hoc network, rooftop-neighbor

network– sensor network

• Better coordination to resolve resource competition from inside the network

Challenges I

• Location-dependent channel contention

• Spatial channel reuse

• Distributed scheduling information

• Notion of fairness

Challenges II

F1

F2

F3X

Spatial Collision

Spatial Channel Reuse

No Spatial Channel Reuse

F4

Flow Contending Graph

F1

F2

F4

F3

F1F4

F2

F3

Flow Contending

GraphNode Graph

RTS-CTS-(DS)-DATA-ACK

Challenges III

• Distributed scheduling information– NO single entity has

complete flow information

• Fairness notion– No consistent

contending flow set– Indirect contention– Conflict with max t’put

F1F4

F2

F3

Flow Contending

Graph

Fairness

• Flow with minimum service should be guaranteed to receive service

– Identify the flow with global minimum service w/o global search? – MLM-FQ

• Simultaneous transmissions should be scheduled whenever possible, subject to max-min and BW constraint

– EMLM-FQ

MLM-FQMaximize Local Min-FQ

• Identify all flows that receive local minimum services

• Global minimum must be among local minima

• Maximize-global-min => maximize-local-min

MLM-FQ

• A node tags its own flows (STFQ)• Piggyback service tags in control

messages:– Current service tag in RTS-CTS– Updated service tag in DS-ACK

• A node maintains one-hop neighboring flows’ tags – a local table

• A node transmits only if one of its flows has the local minimum service tag

MLM-FQ

F1F3

F2

F4F2F4

24

F1 1

F2F4

24

F1 1

F4 4F3 3

F2F3

23

F1 1

F4 4

F1 be scheduled

Drawbacks

• Low channel utilization – spatial reuse may be prohibited– Worst-case C/N

• Deadlock due to collisions on service tag propagation

F1 1F2 2F3 2

F1 1F3 2F4 1

F1 1F2 2F4 1

F2 2F3 2F4 1

F1 11

F4 11

F1 1

F4 1

F1 1

F4 1

Enhanced MLM-FQ

• Set a backoff value for each flow before it contends for channel

• Backoff value set to be the total number of flows that have smaller service tags in local table

How E-MLM works

How E-MLM works

EMLM-FQ Min Fair Service

• Worst-case: C/Nc– Nc is the total

number of flows in the maximum clique of the flow contending graph

– Independent from total number of flows

Comparing Scheduling Schemes

Performance Evaluation

FTP/TCP

FTP/TCPCBR/UDP

Throughputs

Aggregate: 205.8kbps Aggregate: 141.4kbps

SCOPE Summary

• Ensure fair service to individual flows through localized coordination

• Table-driven distributed algorithms• Addresses:

– Location-dependent resource sharing – self-coordination from inside the network

– Scalability - both state maintenance and communication overhead

• Mobility & wireless link dynamics due to outside interferences and attacks