Wireless Sensor Networks: Random, Ephemeral Identifiers

1
Wireless Sensor NetworksRandom, Ephemeral
Identifiers
AND
Jeremy Elson and Deborah Estrin Laboratory for
Embedded Collaborative Systems --
http//lecs.cs.ucla.edu
WSNs create new opportunities...
...while presenting new challenges.

• New technologies have reduced the cost, size, and
  power ofmicro-sensors, micro-actuators (MEMS),
  and wireless interfaces.
• Systems can
• Sense phenomena at close range
• Be embedded in the environment - outdoors,in
  buildings, vehicles, equipment, people?
• Potential revolutions in
• Environmental monitoring
• Disaster relief scenarios
• Condition-based maintenance
• Fantastic Voyage?

• Energy constraints imposed by untethered and
  unattended systems.
• Scaling challenges due to very large numbers of
  sensors.
• Level of dynamics
• Environmental obstacles, weather, terrain,
  changes in RF propagation...
• System ad hoc deployment, node failures,
  self-configuration, re-tasking
• Many assumptions in traditional distributed
  systems are violated
• Passively listening for messages from others is
  very expensive
• Transmission should be avoided even if the
  channel is idle
• Localized algorithms - do computation if it
  avoids transmission
• In-network processing (e.g., summarization), not
  simply routing
• Self-configuration - system should be adaptive,
  robust to dynamics
• Synchronization - ad-hoc creation of local time
  and space coordinates

In parallel with others advancing the state of
the art in hardware, our lab studies techniques
for scalable, energy-conservative, robust
coordinationhow do you get all those sensors to
do something useful?
Unique Identifiers A staple of distributed
systems, but do they apply here?
Necessary but Too Expensive
Is there another way?

• Sensor nets have many uses for unique
  identifiers(packet fragmentation, reinforcement,
  compression codebooks...)
• But, its critical to maximize usefulness of
  every bit transmitted and received each one
  brings the net closer to its death (Pottie)
• Low data rates mean no space to amortize the cost
  of a large, guaranteed unique ID -- e.g.,
  statically assigned during manufacture, ala
  Ethernet
• High dynamics mean the cost of a protocol to
  assign small locally unique IDs by resolving
  conflicts cant be amortized over lifetime of the
  network

• Can we use a series of small, random, ephemeral
  identifiers instead?
• In case of an identifier collision do nothing!
  Avoid persistent collisions by picking a new ID
  for every transaction
• Marginal cost of losses from ID collisions will
  likely be small compared to losses from many
  other sources (packet loss, node dynamics, etc.)
• Sensor nets must be robust to dynamics anyway -
  so, perhaps this small additional source of error
  does not cost us anything

Example Packet Fragmentation

• Address-free fragmentation using random
  identifiers
• Sender picks a random number for each new
  packet,allowing receivers demultiplex
  concurrently arriving packets
• Identifier collisions result in lost packets
  (e.g. failed checksum)
• Scheme scales beautifully identifier space grows
  withtransaction density, not network size, node
  density, etc.
• Leverages spatial and temporal locality (see
  diagram at right)
• Requires no configuration
• Random IDs will be IDsif total net size is large and transaction
  density small
• Can promote good design, making dynamics the norm

A model of the performance of random identifiers,
validated by experiments

• Definition of efficiency
• Total bits transmitted Address or Identifier
  Bits Data Bits
• For static allocation Bits received Data Bits
  Only
• For AFF allocation Bits received Data Size
  P(No-Collision)
• P(N-C) (1 - 1/IDspace) (trans-density)
  IDspace 2address-bits

• Implementation on our SCADDS testbed
• Low-power Radiometrix RPC radio (27 byte
  frames)
• PC104 stacks (just like a PC)
• Test of address-free fragmentation
• Give each node a unique ID, report loss due to ID
  collisions
• Try a listening heuristic to avoid collisions

3
(address bits are not considered useful)