Review possible simulator designs

[dnorman]

@mmacfadden and @dnorman to meet up and work out the design for the network simulator.

Deliverables:

• Decide upon and document a final design for the network simulator.

Considerations:

• Deterministic unit tests
  • How to make threading scenarios deterministic?
  • How to consider deterministic "execution time" for computational tasks
  • How to ergonomically specify Minkowski coordinates for operations
  • Do we want to employ futures, and if so, how?
• Preparation for Monte Carlo simulations
  • Design sufficiency
  • document initial scenario to validate subsequent implementation
  • how to log different orderings of events which occurred during monte carlo runs
  • how to measure performance sufficient to provide scientific data

[dnorman]

Meeting findings:

1. Simulator probably ought not take a "fate" based approach, but rather a "ray-tracing" or conditional based approach. This will allow for substantially more flexibility in terms of introducing artificial latency, failures, relativistic effects, AND the inclusion of application state conditional checking (not very efficient, but may be useful in some cases)
2. Simulator should schedule both Memo delivery AND execution of closures, as necessary to service the needs of behavioral simulators in a manner which is consistent with the global ordering of events.
3. Simulator shall allow scheduled events to be canceled, so as to allow for cases in which scheduled durations of processes are inappropriate given new information observed prior to the scheduled time ( presumably triggered by another party's advancement of the clock )

Proposed SystemSimulator API:

simulator.schedule(500,||{ ... }) thread_handle = simulator.block_until_event(||{ ... })