Refactor benchmarks

[Jake-Shadle]

This refactors the benchmarks from using criterion to divan. Divan is far more lightweight in general, and should make benchmarks easier to add in the future versus criterion. There are some missing features like HTML reports (planned?) and perf diffing (maybe?) from last runs, but IMO those are okay to lose for now in favor of easier benchmarking.

Here is the current output:

Timer precision: 20 ns
read_write     fastest       │ slowest       │ median        │ mean          │ samples │ iters
├─ read                      │               │               │               │         │
│  ├─ direct                 │               │               │               │         │
│  │  ├─ 254   51.63 ms      │ 56.87 ms      │ 54.57 ms      │ 54.47 ms      │ 10      │ 10
│  │  │        49.18 MB/s    │ 44.65 MB/s    │ 46.54 MB/s    │ 46.62 MB/s    │         │
│  │  ├─ 508   54.71 ms      │ 57.86 ms      │ 56.47 ms      │ 56.39 ms      │ 10      │ 10
│  │  │        92.84 MB/s    │ 87.79 MB/s    │ 89.94 MB/s    │ 90.07 MB/s    │         │
│  │  ╰─ 1500  56.2 ms       │ 58.71 ms      │ 57.66 ms      │ 57.6 ms       │ 10      │ 10
│  │           266.8 MB/s    │ 255.4 MB/s    │ 260.1 MB/s    │ 260.3 MB/s    │         │
│  ╰─ quilkin                │               │               │               │         │
│     ├─ 254   117.7 ms      │ 147.6 ms      │ 118.3 ms      │ 121.5 ms      │ 10      │ 10
│     │        21.56 MB/s    │ 17.2 MB/s     │ 21.45 MB/s    │ 20.9 MB/s     │         │
│     ├─ 508   115.7 ms      │ 123.4 ms      │ 122.7 ms      │ 121.4 ms      │ 10      │ 10
│     │        43.9 MB/s     │ 41.15 MB/s    │ 41.38 MB/s    │ 41.83 MB/s    │         │
│     ╰─ 1500  121.1 ms      │ 130.8 ms      │ 124.1 ms      │ 124.6 ms      │ 10      │ 10
│              123.7 MB/s    │ 114.6 MB/s    │ 120.8 MB/s    │ 120.3 MB/s    │         │
╰─ write                     │               │               │               │         │
   ├─ direct                 │               │               │               │         │
   │  ├─ 254   51.24 ms      │ 60.71 ms      │ 54.83 ms      │ 55.22 ms      │ 10      │ 10
   │  │        49.57 MB/s    │ 41.83 MB/s    │ 46.32 MB/s    │ 45.99 MB/s    │         │
   │  ├─ 508   50.65 ms      │ 58.14 ms      │ 55.77 ms      │ 55.27 ms      │ 10      │ 10
   │  │        100.2 MB/s    │ 87.36 MB/s    │ 91.07 MB/s    │ 91.9 MB/s     │         │
   │  ╰─ 1500  56.42 ms      │ 58.23 ms      │ 57.12 ms      │ 57.19 ms      │ 10      │ 10
   │           265.8 MB/s    │ 257.5 MB/s    │ 262.6 MB/s    │ 262.2 MB/s    │         │
   ╰─ quilkin                │               │               │               │         │
      ├─ 254   115.7 ms      │ 119.8 ms      │ 119.3 ms      │ 118.5 ms      │ 10      │ 10
      │        21.93 MB/s    │ 21.18 MB/s    │ 21.28 MB/s    │ 21.42 MB/s    │         │
      ├─ 508   120 ms        │ 124.4 ms      │ 120.9 ms      │ 121.3 ms      │ 10      │ 10
      │        42.32 MB/s    │ 40.83 MB/s    │ 41.98 MB/s    │ 41.84 MB/s    │         │
      ╰─ 1500  123.7 ms      │ 132.6 ms      │ 124.6 ms      │ 125.7 ms      │ 10      │ 10
               121.2 MB/s    │ 113 MB/s      │ 120.2 MB/s    │ 119.3 MB/s    │         │

[Jake-Shadle]

Note there is a slight bug somewhere where starting a quilkin proxy in the benchmarks will spin up, but no socket writes will reach it. I think this might be some form of leak/resource exhaustion since starting a new benchmark process will "fix" the issue. I'm fairly sure this issue exists on main, but it's not noticed since the current benchmarks only spinup 2 instances for all of the benchmarks, whereas these spinup a new instance for each package size on both read and write.

[quilkin-bot]

Build Succeeded 🥳

Build Id: 0850f05c-e64f-4226-a170-016ca3d78420

The following development images have been built, and will exist for the next 30 days:

• us-docker.pkg.dev/quilkin/ci/quilkin:0.8.0-dev-5d71f5d

To build this version:

git fetch git@github.com:googleforgames/quilkin.git pull/853/head:pr_853 && git checkout pr_853
cargo build