Flyspec

1
Flyspec a network for silicon fly sensors

• David L. Mills
• University of Delaware
• http//www.eecis.udel.edu/mills
• mailtomills_at_udel.edu

2
Assumptions

• Designed for ultra-simple ad-hoc sensor nets with
  short range burst radios
• Simple undisciplined ALOHA medium access without
  carrier sense or backoff.
• Transmissions consist of omnidirectional bursts
  with limited range.
• Entities are of two types
• End nodes (sources and sinks)
• Intermediate nodes (repeaters)
• All nodes must be able to transmit, some sources
  might not be able to receive.
• There is no routing table, no routing algorithm
  and no congestion control.
• Routing principles are based on dynamic
  reverse-path forwarding.

3
Routing principles

• Basic routing principle is to determine when a
  burst for a designated sink is received, should
  it be repeated or not.
• Sources and sinks are assigned distinct IDs
  repeaters have no IDs.
• Every burst sent carries the source ID, sink ID,
  sequence number, which increments for each burst,
  and hop count, which is initialized at zero and
  increments at each repeater.
• When a burst is received, buffer it and
  initialize a counter, which then decrements at a
  fixed rate.
• If the counter value falls below a specified
  threshold, increment the hop count and repeat the
  burst.
• If a burst is received with the same source ID,
  sink ID, sequence number and
• greater than the hop count, purge the buffer.
• equal or less than the hop count, ignore the
  burst.

4
Minding the thresholds

• The basic idea is to delay an appropriate time to
  allow some other node closer to the destination
  to repeat a burst without wasting bandwidth for
  multiple transmissions.
• Each repeater has a list of recently heard burst
  source and sink IDs, together with hop count and
  threshold. Initially, the threshold is set
  depending on the hop count from the source.
• The threshold degrades slowly so to allow
  reconfiguration should a repeater be lost.
• If a packet is received with destination ID not
  on the list, the threshold is set relatively low.
• The repeater will hold on to the burst waiting a
  relatively long time to give other repeaters with
  knowledge time to transmit first.
• A nearby repeater with knowledge will repeat it
  at greater hop count, which will kill the buffer.

5
Questions

• We must assume bursts flow both ways between
  source and sink in order to learn the reverse
  path. So, sinks might run a low-level whisper
  campaign to the sources.
• When the network launches, repeaters dont know
  anything, so bursts will random-walk until
  finding sinks, all the time leaving a trails
  behind where they have been.
• Is this thing stable? Dies it eventually converge
  to shortest paths?
• What are the failure/recovery dynamics? Does it
  scale?