scratch/mevenson/simulation/simulation.org

• Glacier et. al. Simulation
  • Model
    • sync v. async
  • Implementation
    • Rust <https://github.com/logos-co/consensus-prototypes/> (Currently Private to Logos)
• Graphs for @alexanderw
  • Grind out parametrized settings with existing code
  • Double check that this is all we need.
• Make parameters depend on network size
• <2022-09-02 Fri> Initial implementation
  • encode/decode parameters in simulation input/output filenames
    • record random seed used for execution
  • write basic execution pipeline
  • benchmark
  • setup/run batch jobs
    • For @AlexanderM's parameter choices
    • For RFC 38
  • instrument simulation to emit line data
  • Produce graphs
• Colophon

Glacier et. al. Simulation

Model

sync v. async

Implementation

Rust <https://github.com/logos-co/consensus-prototypes/> (Currently Private to Logos)

TODO Graphs for @alexanderw

For given choice of parameters,

	Parameter	Alexander	RFC 38
k	neighbors in query	100 (fixed)	7 (undecided grow to a max 2^4)
l	look ahead	~500	19
a1	alpha1	4/5	4/5
a2	alpha2	1/2	2/5

retain a vector for all executed rounds contain the following three integers per round:

• nodes in round
• nodes finalized to yes
• nodes finalized to no

to produce a plot of total yes, no, and undecided per round.

Simulate up to 10^4 nodes adding 10, 100, 1000?

./target/release-opt/consensus-simulations --input-settings etc/glacier.json --output-file=./var/glauber.output

run Glacier simulator

TODO Grind out parametrized settings with existing code

name	JSON path	Notes
a1	$.consensus_settings.glacier.evidence_alpha
a2	$.consensus_settings.glacier.evidence_alpha_2
l	$.consensus_settings.glacier.look_ahead
k	$.consensus_settings.query.initial_query_size
k_query_size	$.consensus_settings.query.query_size	???
k_multiplier	$.consensus_settings.query_multiplier
k_multiplier_max	$.consensus_settings.max_multiplier
n	$.byzantine_settings.total_size
honest	$.byzantine_settings.honest
yes	$.distribution.yes
no	$.distribution.no

use these on a base template

TODO Double check that this is all we need.

opinion votes can be derived…

TODO Make parameters depend on network size

<2022-09-02 Fri> Initial implementation

TODO encode/decode parameters in simulation input/output filenames

All are encoded except for query-size. What is the difference between this and initial_query_size?

TODO record random seed used for execution

Probably should be able to be specified in configuration as otherwise it needs to be parsed from consensus-simulations stdout.

DONE write basic execution pipeline

CLOSED: [2022-09-04 Sun 08:35]

• CLOSING NOTE [2022-09-04 Sun 08:35]

TODO benchmark

Along machine, architecture, and async/sync axes

TODO setup/run batch jobs

For @AlexanderM's parameter choices

For RFC 38

TODO instrument simulation to emit line data

DONE Produce graphs

CLOSED: [2022-09-04 Sun 08:35]

• CLOSING NOTE [2022-09-04 Sun 08:35]

For given parameter choices show YES, NO, NONE opinion evolution over rounds

Colophon

    @easye <mark.evenson@status.im>
    Created: 01-SEP-2022
    Revised: <2022-09-04 Sun 08:35>