Congestion Control and Fairness

1
Congestion Control and Fairness for Many-to-One
Routing in Sensor Networks
Cheng Tien Ee Ruzena Bajcsy
Background
Mechanism 2 Determining Effective Data
Transmission Rate

• sensor motes send information to 1 central mote
  (or base station)
• since motes may not be within radio range of one
  another, intermediate motes need to route
  packets, resulting in multi-hoping networks
• focus on applications that require gathering of
  data that cannot be aggregated within network

• basically monitor and calculate average time
  taken to send 1 data packet (from first attempt
  to actual transmission)
• why not use known channel rate?
• need to take into account interference when
  neighbors transmit simultaneously
• when interference occurs, effective rate is less
• rate is appended to data packet, children nodes
  eavesdrop on packet and updates

Fairness

• within a period (or epoch), each mote transmits a
  multiple of packets from each subtree equal to
  the size of that subtree
• e.g. A transmits 2 packets from B, and 3 packets
  from C, 1 packet from itself, within 1 period
• requires
• per child queue
• FIFO queues
• subtree size (obtained as before)
• correctness proof by induction

Motivation

• current protocols do not scale to more than 10s
  of nodes!
• motes generate more data than can be sent to base
  station (problem 1 congestion)
• packets get dropped, waste of precious energy
• base station receives more packets from motes
  (problem 2 unfair)
• need to tell nodes to send at particular rate,
  and consider fairness
• easy to extend solution to multiple-base stations
  scenarios
• also extends to scenarios where a subset of motes
  generates data

Bs queue
Cs queue
As queue
B
C
A
F
A
D
F
E
B
D
C
Congestion Control
Simulation Implementation Results

• first, differentiate between data generation rate
  and effective transmission rate
• data generation rate rate at which application
  generates data
• effective transmission rate effective rate at
  which mote transmits data, which may include data
  from downstream motes
• next, require 2 mechanisms at each node
• should know the number of downstream motes, in
  other words, children in its subtree
• should know the local effective transmission rate

• need simulations to check performance for huge
  network (hundreds to thousands of nodes)
• simulated
• radio transmission rate (19.2kbps)
• MAC protocol (Multiple Access Collision
  Avoidance, MACA)
• RTS/CTS packets of size 2 bytes each
• data packets of size 30 bytes
• packet loss due to interference and queue
  overflow
• implemented
• in 10 Mica2dot motes
• MAC protocol is MACA, with hop-by-hop ARQ

• if parent node has lower generation rate than
  this node, use parents rate,
• else rate of all children in subtree
• e.g. if node As transmission rate is 70
  pkts/sec, then any node in its subtree can
  generate at most 10 pkts/sec
• disseminate control information piggy-backed on
  data packets

Conclusion

• network adapts itself automatically
• exact same, simple code runs in each mote
• very scalable
• queue size can be small and constant
• state grows linearly with number of neighbors
• congestion control and fairness can be
  implemented in transport layer, little or no
  modifications need to be done to MAC layer

Mechanism 1 Subtree Children

• each node sends children 1 to its parent
• done recursively towards the base station
• can adapt to topology changes easily
• eg node C sends 1, node B sends 3, node A
  sends 4