Paper Outline:

• Introduction
  • Many mobile network devices with multiple interfaces
  • P2P opportunities often left untapped
  • Multiple interfaces introduces problem of interface selection
  • Given different power consumption metrics, different choices can be made
  • Goal of paper: how to characterize power consumption (measurement based)
  • Goal of paper: interface scheduling system based on characterizations
    • Tested via simulation
• Characterizing Power
  • Effective Bandwidth
  • Influencing factors at different layers
  • Energy consumption
  • Metric: efficacy (bytes/joule)
  • Assumes equal utility on all bytes
  • Generalize sum(byte*utility/joule)
    • Generalized metric not used in this work, but still introduced
• Applying metric: measurements
  • Experiment group 1: assuming no sleeping. Only looks at TX/RX differentials
  • Experiment group 2: sleeping. looks at full tx/rx differentials
  • observations from experiments.
  • use conclusions to motivate design decisions in following section
• Scheduling system:
  • Observation: ``Bandwidth density`` is key metric
  • Constributions: interactivity, channel quality
  • How can we guage interactivity?
    • Packet counting/averaging
    • Known port number behaviors
  • How can we guage channel quality?
    • PHY-layer feedback
    • Acheived bandwidth vs. expected bandwidth
  • Local-proxy based daemon manages all interfaces, and schedules transmissions

over interfaces based on expected application demand and estimated channel quality

• System evaluation
  • Simulation with parameters based on measurements
  • Different interactivity profiles and channel efficiencies
  • Results: energy savings with system based on simply sticking with one

interface

• Measured: Time for transcation and power consumption

• Conclusion
  • Using efficacy metric + application behavior to effect interface scheduling
  • Future work: including tighter feedback control loop with lower layers
  • Future work: weighting data differently in efficacy metric