Load testing on reference hardware

[Voxelot]

Given an upper time constraint on block execution (eg 2s), devise a framework determine what the max gas per block should be to fit that target on any given hardware configuration.

Additionally, throughput test a variety of different transactions to evaluate our performance target. These could be basic utxo transfers vs swayswap exchanges.

Deliverables:

• [ ] Regular reports of gas prices and some way to tell if an opcode's cost has dramatically changed. Benchmarks, analyze the benchmarks, sanity checks. These can be separate chron-jobs.

[freesig]

Let's use the same same as we use in beta and do the benchmark many times and take an average.

[freesig]

just bench noop

[freesig]

• [ ] Duplicate existing production docker image
• [ ] Modify to docker to build benchmark bin

[xgreenx]

Setup GitHub action to run it, replace with dedicated instance later

[Voxelot]

Ideally the load test would utilize the SDK to give us an indicator of full e2e performance in our stack, so we can identify bottlenecks that would affect actual users.

[MitchTurner]

Is this a feature?

[xgreenx]

It is related to devops and CI more. The main idea is to automate benchmark result generation. Right now it is done manually.