This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Craft learn about neighbours over the link. Would late information make each craft pass on more of a disturbance than it receives, so that it grows instead of dying out?
What was tested
The full rules ran on the shoved hovering grid with 2 messages a second and 5 % of messages lost. Link delay went from 0 to 2 s.
What was found then
The share of the shove each craft passed on rose from 0.84 with no delay to 0.93 at 2 s. That is closer to 1, where disturbances would start to grow, but never across it in this setup. Delay wears the damping down; here it did not break it.
What changed
Nothing.
Numbers and source references
- 2 broadcasts a second
- link rateexperiments/run_experiments.py@a2cdb52 (f_tx=2.0)
- 5 %
- packet lossexperiments/run_experiments.py@a2cdb52 (p_loss=0.05)
- 0 to 2 s
- delay sweptresults/results.json e3_latency_sweep
- 0.84
- α at no delay (0.843)results/results.json e3_latency_sweep
- 0.93
- α at 2 s delay (0.927)results/results.json e3_latency_sweep