Empirical study of large scale network dynamics joint work with Alec ucb, Bhaskar cornell

1
Empirical study of large scale network dynamics
joint work with Alec (ucb), Bhaskar (cornell)

• 170 node experiments on understanding performance
  of large-scale algorithms
• Isolate density, connectivity-based effects and
  -based effects
• Define better metrics to define connectivity
  radius, asymmetric links
• Focus on pathological instances
• To what extent does asymmetry affect protocol
  design?
• How do collisions manifest themselves in
  flooding-based tree-building algorithms
• Why is there high-clustering behavior in trees
  constructed from flooding?

2
Impact of Asymmetric Links

• Between 10-20 asymmetric nature of links
  observed depending on transmit signal strength of
  nodes.
• Asymmetric links form a significant fraction of
  long links
• Why is this significant?
• Reverse-shortest path algorithms tend to favor
  long-links, that exhibit higher percentage
  asymmetry.

3
Future Work Tuning Network Structure to
Aggregation Function

• At network level, understand parameters that will
  help construct different tree structures
• Different aggregation functions have different
  properties
• Sparse/Dense data distribution
• Aggregatability what kind of transformations can
  be performed on data
• Frequency of data generation, size of data.
• Exploit domain knowledge to build tree
• Obtain appropriate datasets