Service Allocation Simulator

Lessons:

• Simple Hill-Climbing scenario where agents tell sensors to look their way and sensors obey them (even if it means forgetting about previous requests:
  • In some runs, it never converges.
  • Plotting t,t+1 in scatter plot shows that those runs that achieve high utility do not diverge as much. The ones in the middle tend to fluctuate +-10.
  • The system (on average of 100 runs) mostly stablizes after 300 steps. At this point the system has achieved a self-organized criticality, so we can expect some avalanches of changes with dynamics obeying a power-law distribution. Experiments show that longer changes still appear at later times, but we cannot confirm a power-law distribution. (probably due to lack of locality).
• Bidding Agents tell the Bidding services their marginal utilities. Every so often the services clear and perform the service with highest agregate demand.
  • They get higher utility, around 200 better after the first clearing, on average.
  • They do not stabilize as much, but their fluctuations (deviations) have smaller magnitude.