Congestion Control in CSMA-Based Networks with Inconsistent Channel State

1
Congestion Control in CSMA-Based Networks with
Inconsistent Channel State

• V. Gambiroza and E. Knightly

Rice Networks Group http//www.ece.rice.edu/networ
ks
2
MotivationCongestion Indication

• Loss

Loss is not a good congestion indicator in
CSMA-based networks
3
MotivationCongestion Indication

• Loss (TCP)

• Buffer occupancy

Buffer Threshold
4
MotivationCongestion Indication

• Loss (TCP)

• Buffer occupancy

Dropped packet
Buffer Threshold
Distributed queue
5
Our Approach

• Assumptions
• Unmodified MAC such as IEEE 802.11
• Decoupled vs. joint design
• Utility maximization congestion control
• Define and incorporate key issues and challenges
• Study their impact on performance

6
Outline

• Key issues and challenges
• Background
• Utility function
• Utility maximization congestion control
• Results
• Inconsistent states
• Comparison with TCP

7
Issues and Challenges in CSMA-Based Networks

• Channel state in multihop networks
• Inconsistent

8
Issues and Challenges in CSMA-Based Networks

• Channel state in multihop networks
• Inconsistent

• Data transmission capacity
• Actual capacity for data transmission unknown
• Depends on number of competing flows, node
  locations, propagation environment
• Efficiency Index ?
• Fraction of C available for data transmission

• Distributed queue service order
• Service order is not FIFO
• Information asymmetry service order (close to)
  strict priority

• State observation and sharing
• Multiple metrics measured

9
Issues and Challenges in CSMA-Based Networks

• Channel state in multihop networks
• Inconsistent

• Data transmission capacity
• Actual capacity for data transmission unknown
• Depends on number of competing flows, node
  locations, propagation environment
• Efficiency Index ?
• Fraction of C available for data transmission

Critical to performance
Not incorporated by any of the prior work

• Distributed queue service order
• Service order is not FIFO
• Information asymmetry service order (close to)
  strict priority

• State observation and sharing
• Multiple metrics measured

10
Utility Function

• Degree of users satisfaction with service
  quality
• Quality indicators bandwidth, time (delay),
  power
• Bandwidth utility function
• Relation between users satisfaction and network
  bandwidth

11
Utility Function(Example)

• Voice traffic

Utility
ß
Bandwidth
12
Utility Function(Example)

• File transfer U S a t
• S users happiness when transfer is infinitely
  fast
• t transfer time
• a rate at which satisfaction decreases with time

Utility
S
S/a
Time
13
Utility Maximization Congestion Control
Assumes knowledge of utility functions

• C - capacity vector
• A routing matrix
• x vector of allocated rates (xr allocated
  rate of user r)
• Ur(xr) utility
• Increasing, strictly concave, continuously
  differentiable, additive

14
Congestion Control Algorithm
Approximation
U(x) logx
µj(t) penalty function, price charged by
resource j

• For any initial condition congestion control
  algorithm converges to the unique solution
• Sum of the prices needs to be known
• Price can be conveyed using just one bit feedback

15
Example

• Node B unaware of D-E transmission
• Bs perception of congestion and feedback
  incorrect

Impact on performance?
16
Our Approach

• Assumptions
• Unmodified MAC such as IEEE 802.11
• Decoupled vs. joint design
• Utility maximization congestion control
• Define and incorporate key issues and challenges
• Study their impact on performance
• Single hop topologies
• Possibly a part of more complex multihop
  scenarios
• No collaboration
• Consistent states
• Inconsistent states
• Collaboration
• Measurement metric
• Comparison with TCP

17
Our Approach

• Assumptions
• Unmodified MAC such as IEEE 802.11
• Decoupled vs. joint design
• Utility maximization congestion control
• Define and incorporate key issues and challenges
• Study their impact on performance
• Single hop topologies
• Possibly a part of more complex multihop
  scenarios
• No collaboration
• Consistent states
• Inconsistent states
• Collaboration
• Measurement metric
• Comparison with TCP

18
Congestion Control with Inconsistent States and
w/o Collaboration
Different transmission and carrier sense ranges
Leads to inconsistent states
19
Difference in Channel States
20
Throughput and Fairness PropertiesDifferent
Transmission and Carrier Sense Ranges

• Convergence to unfair rates

? 0.8
21
Congestion Control with Collaboration

• Collaboration
• Nodes collaborate in order to realize true
  channel state
• Study effects of collaboration
• Study choice of measurement metric

22
Congestion Control with CollaborationResults

• TCP Has 2 outcomes

23
Congestion Control with CollaborationResults

• TCP Has 2 outcomes
• UMCC Fair shares very close to ideal rates
• UMCC achieves throughput up to 17 higher than TCP

24
Conclusions

• Framework to study key issues in CSMA-based
  networks
• Channel state, data transmission capacity,
  service order, state observation and sharing
• No globally optimal data transmission capacity
  even with consistent states
• Inconsistent states lead to convergence to unfair
  rates
• Collaboration among nodes alleviates the problem
• Per-flow measurement needed
• Compared to TCP starvation removed, better
  fairness, 17 higher throughput

25
Congestion Control in CSMA-Based Networks with
Inconsistent Channel State

• V. Gambiroza and E. Knightly

Rice Networks Group http//www.ece.rice.edu/networ
ks