SpeckZone Formation and Maintenance in Specknets

1
SpeckZone Formation and Maintenance in Specknets

• Andrew Bates
• c.a.bates_at_sms.ed.ac.uk

2
Overview

• Why do we need speckZones?
• The algorithm to form and maintain them.
• What is a good speckZone?
• Size
• Shape
• Connectivity
• Density
• Controlling the size
• Determining the number of specks in a zone.

3
The need for speckZones

• An individual speck does not have enough memory
  or processing power to be useful.
• We can only run very simple programs in a few
  Kilobytes of memory.
• It is impossible to route between thousands of
  individual specks with the current memory
  constraint.
• An individual speck moves around, is unreliable
  and spends a lot of its time sleeping.

4
The need for speckZones

• Group specks into speckZones of roughly equal
  size.
• The blue internal specks share memory and
  processing power.
• The green border specks handle communication with
  neighboring zones.
• Each speck can only be a member of one speckZone.

5
The need for speckZones

• We can now program a speckZone rather than an
  individual speck.
• This gives us increased memory, processing power
  and reliability.
• We only have to route messages between a few
  hundred zones rather than a few thousand
  individual specks.

6
Constraints for the speckZone algorithm

• The algorithm should be fully decentralised.
• It should not rely on a single speck to manage
  each speckZone, as specks are unreliable.
• Synchronisation should not be required.
• The speckZone must continue to operate if some of
  its specks move away or run out of power.
• The size of the speckZone must be maintained
  despite the unreliability and mobility of specks.

7
The algorithm

• The algorithm is decentralised, leaderless and
  uses only local information.
• Each speck has a state (or colour) that is
  initially random.
• Specks exchange colour information with their
  one-hop neighbours.
• If more than half of the surrounding specks have
  a different colour from a given speck, that speck
  will change colour.
• The new colour is chosen according to the
  distribution of colour in the local area.
• Specks like to be the same colour as most of
  their neighbours.

8
Simulation (sparse network)
9
Simulation (dense network)
10
A good speckZone

• SpeckZones will have an optimum size
• Large zones have more memory/processing power.
• But if the zone is too large, it will be
  expensive to manage.
• The members of a zone should be close together
  to reduce power required for communications
  within the zone.
• Applications that require a large amount of
  communication between the zone may benefit from a
  large surface area along zone boundaries.

11
Controlling the size of a speckZone

• As we have just seen, there is a relationship
  between the density of the network and the
  natural size of the zones.
• The density of the network determines the average
  number of one-hop neighbours.
• Using position information from Ryans Algorithm,
  we can control the number of one-hop neighbours
  and therefore the size of the zone.

12
Controlling the size of a speckZone

• Specks start with a communication range of zero.
• If the zone size is below the specified minimum,
  the communication range is increased.
• If the zone size is above the specified maximum,
  the communication radius is reduced.
• The amount of alteration is proportional to the
  difference between the required and actual zone
  size.

13
Preventing zones becoming too large

• Before a speck changes state, it must check if it
  will make the new zone too large. If so, it
  should not change state.
• If we have a maximum zone size of 20, the blue
  speck must not change to green, as this would
  create one large zone.
• If the zone is too large, the specks on the edge
  of the zone should change state with an increased
  probability.

14
Counting the number of specks in a zone

• Each speck now needs to know how many specks are
  in its current zone.
• A zone must be able to count itself in a
  decentralised way.
• Each speck maintains a list of the members of the
  zone that it belongs to.
• This must be updated as the specks moves around
  or join different zones.

15
The simple counting method

• Each speck periodically sends a message
  containing its ID and a sequence number.
• Specks store these ID, sequence number pairs
  and broadcast them on every iteration of the
  algorithm.
• Any specks that have not been heard from for a
  few iterations are removed from the zone count.
• A disadvantage of this approach is that messages
  must be sent when there are no changes in the
  zone.

16
The efficient counting method

• Only changes to the members of a zone are
  broadcast throughout the zone.
• This significantly reduces the amount of
  information that is sent in static networks.
• If a new speck joins a zone, a message saying
  that it has been added propagates throughout the
  zone.
• When a speck leaves the zone, a message saying
  that it has left propagates throughout the zone.

17
Joining and leaving zones

• When a speck changes state, it must
• Inform the old zone that it has left (if
  possible)
• Inform the new zone that it has joined
• If a speck does not hear from a neighbouring
  speck for a few iterations, it should broadcast a
  message advising that the neighbouring speck has
  left the zone.
• If a speck receives a message saying that it
  should be removed from the zone, it must still be
  in the zone.

18
Maintaining accuracy when specks leave a zone

• A single speck leaving the zone may cut off more
  than one speck.
• Consider speck A. The source of B, C and D is B.
• If B leaves the zone, B should be removed form
  As zone members list.

• A will now over-estimate the zone size at 3, as
  it thinks C and D are still in the zone.
• When B is removed, any other specks whos only
  source is B should also be removed (C and D in
  this case).

19
Simulation (controlling the zone size)

• 1,000 Specks
• 100 Iterations
• Variable communication range
• Variable iteration times
• Minimum zone size 30
• Maximum zone size 40
• Using the simple counting method, due to a lot of
  movement in the Specknet.

20
Simulation (controlling the zone size)
21
Simulation results
Sim time