InterferenceAware Fair Control in Wireless Sensor Networks

1
Interference-Aware Fair Control in Wireless
Sensor Networks

• Present by Zhe Zhou

2
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

3
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

4
Introduction

• We need congestion control in wireless sensor
  network
• Structural Health Monitoring
• Flat sensor network for low-rate periodic sensing
• Tiered sensor network for high data-rate
  applications complicated topology makes
  congestion control more tricky

5
Introduction

• How to ensure fair and efficient transmission
  rates for each nodes in a sensor network?
• Interference-Aware Fair Rate Control (IFRC)
• Transport layer, based on CSMA and routing layer
  (link quality based path selection)
• Distributed
• Use average queue length to detect congestion
• Low-overhead congestion sharing
• Signals all related nodes
• Use AIMD to converge to fairness

6
Introduction

• The challenge
• Hard to determine the related nodes
• Hard to rapidly signal them

7
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

8
Related Work

• TCP Congestion Control
• AQM (Active Queue Management)
• TCP for ad-hoc wireless networks
• Extension of RED on wireless networks
• Congestion mitigation and congestion control

9
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

10
Motivation and Definitions

• r16r20r21

• r20

• r21

11
Motivation and Definitions

• Assumptions
• TinyOS
• CSMA (Carrier Sense Multiple Access) and RTS/CTS(
  Request to Send / Clear to Send)
• Token-Based and TDMA MACs are not considered
• Static Routing Tree in most experiments
• IFRC can adapt to changes in routing tree
• IFRC achieves higher overall throughput on
  routing protocols based on link-quality merics

12
Motivation and Definitions

• Assumptions (continued)
• Link-Layer Retransmissions
• IFRC performs well when link-layer
  retransmissions recover from most packet losses
• Impact of packet losses will be described later
• Definitions
• Fair and efficient
• Each flow fairly divides the channel capacity
• IFRC Each flow receives at least the most
  congested fair share rate
• Not absolutely fair Flows having fewer
  contenders can send at a higher rate to ensure
  overall efficiency

13
Motivation and Definitions

• Definitions (continued)
• Interfering Links
• A link l1 interferes with a link l2 if a
  transmission along l1 prevents the initiation or
  the successful reception of a transmission along
  l2
• Potential Interferer
• A node n1 is a potential interferer of node n2 if
  a flow originating from node n1 uses a link that
  interferes with the link between n2 and its parent

14
Motivation and Definitions
15
Motivation and Definitions

• In tree-based communication, the potential
  interferer of a node include
• Its subtree
• Its neighbor and parents subtree
• Its parents neighbors subtree
• Definition (again!)
• Fi Set of flows routed through node i,
  including flows originating at i and its subtree

16
Motivation and Definitions

• Definition (continued)
• B Nominal total bandwidth
• Fi Fi Fj , j is either a neighbor of i, or a
  neighbor of is parent ( set of all potential
  interferers)
• fl,i the assigned rate of each flow in Fi
• fl minimum of all fl,i

17
Motivation and Definitions

• F16 20, 21, 14, 16, 17, 13, 12, 15, 18, 19
• Nodes contribute to the arrival rate of 16 16,
  20, 21
• Nodes contend with 16 others

18
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

19
IFRC Design

• Main Task
• Congestion Detection
• Signaling
• Rate Adaptation
• Congestion Detection
• Channel Utilization
• Queue size
• With a MAC with carrier-sense, backoffs and
  retransmission, overloaded traffic will increase
  the queue size. Therefore, we simply use queue
  size to indicate congestion.

20
IFRC Design

• Congestion Detection (con.)
• EWMA (Exponentially Weighted Moving Average) for
  estimating average queue size
• Updated for each packet inserting
• A node is congested if avgq gt U , and returns to
  uncongested state if avgq lt L
• Sometimes a single halving is not enough. To
  determine if multiple halving should be executed,
  we need multiply U.

21
IFRC Design

• Congestion Detection (con.)
• We define multiply U as below (k is a small
  integer and I is a constant increment of queue
  length)
• So that as k increases, the difference between
  U(k) and U(k1) decreases, resulting in more
  frequent rate halving which accelerates the
  draining of queue.

22
IFRC Design

• Congestion Sharing
• Insert congestion related information in header
  of each outgoing packet
• Current ri and average queue length
• A bit indicating whether any of its children is
  congested
• The smallest rate rl among all its congested
  children and ls average queue length
• To this point, all neighbors of an arbitrary node
  can receive the congestion information of this
  node and the nodes in its subtree.

23
IFRC Design

• Congestion Sharing (con.)
• Two rules for implicitly notify all potential
  interferers
• Childs rate can never surpass parent
• A node will adapt its rate when congestion occurs
  either at its neighbor or the neighbors subtree

24
IFRC Design

• Rate Adaptation
• Average value of ri is not the max rate by which
  i generate traffic
• At the beginning, a node starts its sending rate
  at rinit and add F to its rates every 1/ ri
  seconds.
• The node continues to increase the rate until
  itself congested or the two rules satisfied Then
  it adapts the rate accordingly.
• After the adaptation, the node increases its ri
  by d/ri every 1/ri seconds.

25
IFRC Design

• Base Station Behavior
• Sets the initial rate rb to the nominal rate of
  the channel and do not increases it
• If any of its children is congested, decreases
  its rate, and broadcasts it twice
• After each adaptation, increments rb by d/rb
  every 1/ rb seconds. As the station itself has no
  data to send, it broadcasts its rate after at
  least m packet have been received from the
  fastest child.

26
IFRC Design

• Extension to IFRC
• Multiple Base Stations
• If one of the children of the base station is
  congested, the base station sends a control
  packet indicating that.
• Weight Fairness
• When only a subset of nodes transmit

27
IFRC Design

• Discussion
• IFRC can not implemented over an unreliable MAC
  layer
• IFRC can not detect interference from
  non-neighboring nodes
• IFRC can not work on cards turning off
  overhearing (Battery Killer!)
• IFRC will work when intermediate nodes perform
  in-network aggregation

28
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

29
Parameter Selection In IFRC

• Intensity in AIMD
• Each node i increases its rate ri by d/ri every
  1/ri seconds.
• Namely, it follows a linear curve with slope d.
• For efficiency, d should be as large as possible.
  However, for stability d should be kept not too
  large. So, our task is to find its upper bound in
  terms of maintaining the stability

30
Parameter Selection In IFRC
31
Parameter Selection In IFRC

• To prevent ri ramping from rmin,i to rmax,i in
  one step (in 1/ri seconds), we need d/rmin,i ltlt
  rmin,i , or
• Where 0lt e lt1 is a small positive number. We will
  derive its upper bound below.
• The excess number of packets can be calculated as
• If we focus on one congested node j, and Iij be
  the function that indicates whether packets from
  i traverse j. The total number of excess packets
  could then be denoted as

32
Parameter Selection In IFRC

• Taking the effect of contention into account, we
  substitute Iij with fij.
• We need to tune the value of to validate the
  following two equations
• Equation 1, 2, 3 guarantee system stability and
  only one signal is sent for one node when
  congestion occurs, which mitigates the reduce of
  efficiency.

33
Parameter Selection In IFRC

• By substituting rst in Equation 2 using Equation
  1 and let Fj Si fij, we get

• (See the figure) As rst rises, the difference
  between the area of two triangles increase, thus
  the efficiency decreases.
• As rst drops, the upper bound of edrops, so we
  will get a smallere.

34
Parameter Selection In IFRC

• To prevent a node sending out congestion info in
  the duration of receiving other nodes congestion
  info, we have
• And consequently, we have

Average of si
35
Parameter Selection In IFRC

• So e is restricted by these two equations

• In small network when Fj is small, the first
  inequality determines e.
• In large network when Fj is large, the second
  inequality determines e.
• Use nlogn for Fj (Intuitively, every node
  interferers with j for logn times).
• rst should be something proportional to B/nlogn,
  so we set rinit to B/10nlogn.
• F is set to rinit /8.
• U(0) and U(1) are set to N/2 and N respectively.

36
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

37
Evaluation

• Implementation and Methodology
• 40-node wireless sensor testbed
• TinyOS 1.1 with IFRC plugged in
• Two modules. Neighbors congestion table is
  stored.
• Promiscuous mode enabled, which disables the
  chip-level ack, thus ack in MAC is added.
• Each node Moteiv Tmote with a 8MHz Texas
  Instruments MSP430 microcontroller, 10KB RAM and
  a 2.4GHz IEEE 802.15.4 Chipcon Wireless
  Transceiver with a nominal bit rate of 250 Kbps
• Deployed over 1125 sqare meters of a large office
  floor
• A USB backchannel for logging experiment data
  (which will have some problem later)
• 8 hops, all links have a loss rate lower that
  40, pretty uncomplaining

38
Evaluation
Testbed connectivity graph
39
Evaluation
Window based
Really slow slow-start
Pretty small!
40
Evaluation

• A fixed tree to maintain a same environment for
  all experiments (modifies MultiHopLQI)
• A hour at least for each experiment
• Long experiments, run at usually late at night or
  in early morning
• Every packet transmission, reception, and every
  change in rate at each node (including base
  station, although no transmission) is recorded.

41
Evaluation

• Packet reception ratios range from 66 to 96
• 9 hops deep
• A good topology with all kinds of variance

42
Evaluation

• Every nodes receive approximately fair rates and
  goodput
• Node 13 and 8 are congested (hard to perceive
  from the graph)
• Hop-by-hop recovery resulted in fewer that 8
  packet loss

43
Evaluation

• Instantaneous goodput is stable, with minor
  variations attribute to AIMD.

44
Evaluation

• Nodes adapt their rate nearly synchronically
• Slow start and AIMD is clear visible
• Some nodes adapt their rate slower due to network
  lantency (not shown)
• Horizontal line caused by experiment data loss
  resulted from USB issues

45
Evaluation

• Queue never builds up to higher than 25 -gt no
  packet loss

46
Evaluation

• The efficiency is pretty encouraging.

47
Evaluation
Two successive decreases
48
Evaluation
49
Evaluation
50
Evaluation
51
Evaluation
52
Evaluation
53
Evaluation
54
Outline

• Introduction
• Related Work
• Motivation and Definitions
• IFRC Design
• Parameter Selection In IFRC
• Evaluation
• Conclusions

55
Conclusions

• Conclusion
• IFRC is the first practical interference-aware
  rate control mechanism for WSN
• IFRC is fair
• In terms of efficiency, IFRC is questionable
• Future work
• Implement reliability in IFRC
• A more rigorous proof of the choice of IFRC
  parameters
• A complete analysis of the effects of other
  factors on IFRC

56
Thank you!