Minimal benchmarking framework

[AntoineRondelet]

That would be ideal to have some minimal benchmarking framework. As of today, much benchmarks were carried out manually, but having a better/automated way to run benchmarks would make our life easier and would allow us to better capture the impact of the various PRs on the overall performances of the project.

I'd consider this a first good issue to tackle for anyone willing to get involved in Zeth. A good/dummy first iteration would be to have some timers added to the cpp code if compiled with a BENCHMARK option. That would time remarkable operations for the given set of protocol parameters. The result of this benchmark could then be written in a benchmark folder containing, say, a csv file where each line would correspond to a PR. That way we'd have some basic tracking of the impact of each PR on the perfs of the code, and ultimately track perfs gains/loses across releases. Having this data recorded in a csv would pave the way for basic graphs with python scripts to make the results digestible and easy to share.

This obviously describes some very basic first step for the benchmarking framework, something more elaborate would need to be done later down the line. @dtebbs any thoughts, or things to add?

[AntoineRondelet]

Look into:

• https://github.com/rhysd/github-action-benchmark
• https://github.com/google/benchmark

Ideally, we'd like to benchmark things like:

• The various executables -> for the proof gen time, mpc contribution etc
• The serializarion/deserialization functions and the APIs
• The compilation times
• etc etc

Doing such benchmarks would allow to better track perfs variations across releases and even track down perfs associated to each PR (if the benchmarking tool is incorporated to the CI and ran on-pull-request)