What%20constitutes%20a%20useful%20experimental%20result?

1
What constitutes a useful experimental result?

• Bhaskar Krishnamachari
• Ming Hsieh Department of Electrical Engineering
• USC Viterbi School of Engineering

2
Two kinds of useful experimental results

• Translating theory to practice
• Going beyond current practice

3
Translating Theory to Practice

• From theoretical algorithms to working systems
• Provide Working examples of
• Cross-layer optimization, backpressure
  algorithms, etc.
• Rarely a trivial task
• Do the abstractions hurt performance in practice?
• Do protocols converge despite packet losses,
  random delays? How do they have to be changed to
  make them work in practice?

4
An Example

• The following iterative message-passing algorithm
  can be formally proved to converge to a max-min
  fair rate allocation for a given wireless
  sensor-network data gathering tree

Sridharan, Krishnamachari, 2008.
5
Translating to Practice

• MMF-RC provides max-min fair rate allocation
  under an idealized setting synchronous
  environment, global coordination, static flows,
  perfect links
• In the real world need to figure out how to
  estimate rates, provide distributed operation,
  handle asynchrony, dynamic flows, imperfect links

6

• Solution WRCP, inspired by MMF-RC
• Has a lot of hacks to ensure convergence and
  efficiency empirically no simple analytical
  characterization

7
Going beyond current practice

• Beyond off-the-shelf hardware
• This is the missing link in our field one reason
  for disconnect between physical layer and
  networking researchers

8
Example

• With idealized cooperative flooding (perfect
  channel information, coherent combining), can
  show that the time to get information to all
  nodes is logarithmic in the diameter
• How does this translate to practice?

9
Claim

• We do not have enough principled experimental
  researchers in our field that can talk to,
  understand theorists.