Mitigating Congestion in Wireless Sensor Networks

1
Mitigating Congestion in Wireless Sensor Networks

• Bret Hull Kyle Jamieson Hari Balakrishnan
• (SenSys 2004)
• Presented by Lee, Sehoon
• October 11, 2005

2
Contents

• Introduction
• Congestion Mitigation
• Hop-by-hop flow control
• Rate limiting
• The MAC layer
• Composite approach Fusion
• Experiments
• Conclusions
• Discussion

3
The Problem

• Effective Congestion Control in Sensor Networks
• Distinct characteristics from wired Networks
• Congestion signs buffer drops and increased
  delays in wired Networks
• Increase in interference, poorer channel quality
• Many-to-one data flow can lead to unfairness

4
Impact of Wireless Channel

• A Transmitter can cause interference well beyond
  the TX range
• Greater no. of concurrent transmissions
• Greater error probability

5
Impact of Network Geometry

• Many-to-one communication pattern
• Nodes near the sink act as sources as well as
  relays
• Can lead to unfairness nodes farther away starve
• Packet drops and energy wastage ensues

6
Metrics

• Network efficiency
• ( of useful hops) / (total of transmissions)
• Node imbalance
• ( pkts rcvd at i ) / ( pkts rcvd at is parent
  from i )
• Aggregate sink received throughput
• Network fairness
• Median packet latency
• Buffering / flow control increases delay and
  latency

7
Techniques

• Hop-by-hop flow control
• Rate limiting
• A Prioritized Medium Access Control

8
Hop-by-Hop Flow Control

• Congestion Bit in packet header
• Set by a node that detects congestion
• Nodes that hear packet get feedback
• If parent set congestion bit, stop transmitting
• Congestion Detection
• Queue occupancy
• Queue length increases beyond threshold
• Channel sampling
• Channel busy time estimation

9
Rate Limiting

• Allowed rate determined by estimating no. of
  flows traversing parent
• If N flows, rate is 1/N
• Entails promiscuous hearing
• Token bucket scheme
• To regulate each sensors send rate

10
Token Bucket Traffic Shaping
Tokens generated at rate ?1/N
A (?, s) Traffic Filter Allows traffic at
average rate ? with maximum allowable burst size
s
s
Departing Packet Token
Arriving Packet
A packet may depart only if it can be paired with
a token
11
MAC Techniques

• Prioritized MAC
• Make backoff dependent on local congestion
• More congested nodes choose lower backoff, and
  get priority in channel access
• Shut down RTS/CTS
• Use guard-time to avoid Hidden Terminal problem

12
Application Adaptation

• Helps reduce rate of injection of traffic at the
  original source
• Rate-adaptive applications
• e.g. in sensor networks, might reduce rate of
  periodic sampling, else send data aggregated (and
  compressed) over a window of time

13
Composite Approach Fusion

• All techniques in concert

14
Experimental Evaluation

• 55 node indoor testbed (Mica2)
• Data rate 38.4 Kbps
• Comparison with default TinyOS MAC protocol
• Routing with DSDV to use ETX path metric
• Channel quality aware routes

DSDV Destination-Sequenced Distance-Vector
Routing ETX the Expected number of
Transmissions
16,076 sq. ft. area in of an office building
15
Periodic Workload

• Data generated at fixed intervals
• A sink acts as point of data collection

A typical routing topology
16
Periodic Workload Efficiency
Fusion exhibits best efficiency, even under
increased load
17
Periodic Workload Imbalance
Fusion performs best Most nodes have Imbalance
less than 5
5 nodes (the 90th percentile) have Imbalance
greater than 50
18
Periodic Workload Throughput at Sink
Throughput lower when using Fusion Non-rate
limiting results in higher throughput
19
Periodic Workload Link Loss Rates
Link losses are minimized by Fusion
20
Periodic Workload etc.

• Fairness Latency
• Fairness decreases without congestion control
• Rate-limiting improves fairness
• As load increases, Fusion is the fairest and
  consequently latency is higher

21
High fan-in Workload

• Only a small subset of nodes advertises routes to
  the sink
• Topology has higher fan-in and smaller network
  diameter
• Even at low load, efficiency is lower than in
  Period Workload

Fusion outperforms all strategies at most offered
loads
22
Correlated-event Workload

• Nodes send B packets back-to-back in synchronized
  fashion
• Useful model for detection and tracking
  applications

23
Correlated Event Workload Efficiency
Efficiency best when congestion detection done
using Occupancy Delay
24
Correlated Event WorkloadLatency
Latency is significantly higher when using Fusion
25
Conclusion

• Hop-by-hop flow control with queue occupancy
  improves efficiency for all types of workloads
• Rate-limiting improves fairness
• MAC enhancements support Hop-by-hop flow control
• Fusion dramatically improves network efficiency,
  fairness, and channel loss rates

26
Discussion

• Multiple techniques used in conjunction to
  improve congestion characteristics
• Techniques not new
• Some techniques rely on promiscuous hearing
• Detrimental to power-save schemes