Nix flake

[MathisWellmann]

I found this project after following the issue thread in core-to-core-latency (nviennot/core-to-core-latency#97) about the NUMA issue and decided to give this binary a shot.

This PR adds a flake.nix, which provides reproducible binary packages that can be built with nix build or directly ran with nix run or nix run .#cpu-latency-plot, which is more convenient to me than following x manual steps.

Also convenient to be able to do nix run github:SoilRos/cpu-latency#default directly from github without the need to clone it.

Here are some results for my Zen 5 9654 CPU with SMT enabled.

There was a baseline load average of 5, which explains the outliers.

[SoilRos]

Amazing. Could you perhaps add instructions in the README of how to install it using nix too?

[SoilRos]

And have you had a look why 18th CPU is slower? Maybe the OS was using this thread for something else? What I find even more strange is that the 17th behaves differently than the 18th. AFIK, the only difference between the two is the pipeline.

[MathisWellmann]

I've updated the README with a quick section on Nix builds, directly from the repo.

The 18th CPU is slower due to my cluster workload running websocket trade & quote data processing code at 20-30k msg/s.
I didn't migrate the pods to another host before, so as long as we don't include that run in the README it should be fine for illustrative purposes only, not as ground truth for that CPU performance. Hope that's alright.