AsyncKeyedLock library vs SixLabors.ImageSharp.Web.AsyncKeyLock

[MarkCiliaVincenti]

I am the author of a library called AsyncKeyedLock (NuGet package). I ran some benchmarks comparing my library against SixLabors.ImageSharp.Web.AsyncKeyLock.

|                        Method | NumberOfLocks | Contention | GuidReversals |              Mean |           Error |          StdDev |            Median | Ratio | RatioSD |        Gen0 |       Gen1 |      Gen2 |    Allocated | Alloc Ratio |
|------------------------------ |-------------- |----------- |-------------- |------------------:|----------------:|----------------:|------------------:|------:|--------:|------------:|-----------:|----------:|-------------:|------------:|
|                AsyncKeyedLock |           200 |        100 |             0 |     15,250.916 us |     177.9938 us |     197.8396 us |     15,215.252 us |  1.00 |    0.00 |           - |          - |         - |    4205944 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |        100 |             0 |     18,143.449 us |     358.5022 us |     912.5025 us |     17,996.998 us |  1.20 |    0.06 |           - |          - |         - |    7534456 B |        1.79 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |           200 |        100 |             1 |     35,532.410 us |     445.9032 us |     826.5102 us |     35,266.739 us |  1.00 |    0.00 |           - |          - |         - |    7085744 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |        100 |             1 |     45,359.193 us |     872.9948 us |   1,039.2382 us |     45,268.891 us |  1.26 |    0.04 |           - |          - |         - |   10415272 B |        1.47 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |           200 |        100 |             5 |    127,620.773 us |   2,427.7748 us |   3,779.7529 us |    125,652.856 us |  1.00 |    0.00 |           - |          - |         - |   18606224 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |        100 |             5 |    142,363.037 us |   1,312.9977 us |   1,163.9382 us |    142,018.180 us |  1.11 |    0.04 |   1000.0000 |          - |         - |   21935016 B |        1.18 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |           200 |      10000 |             0 |  1,905,318.441 us |  21,779.0290 us |  20,372.1173 us |  1,902,994.072 us |  1.00 |    0.00 |  21000.0000 | 20000.0000 | 3000.0000 |  398397528 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |      10000 |             0 |  3,129,781.357 us |  61,968.2589 us |  78,370.0061 us |  3,143,483.218 us |  1.64 |    0.05 |  37000.0000 | 36000.0000 | 1000.0000 |  734360424 B |        1.84 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |           200 |      10000 |             1 |  5,083,750.844 us |  62,422.8888 us |  58,390.4090 us |  5,082,294.823 us |  1.00 |    0.00 |  38000.0000 | 37000.0000 | 4000.0000 |  686408568 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |      10000 |             1 |  6,128,064.878 us | 111,756.1418 us | 104,536.7644 us |  6,110,852.719 us |  1.21 |    0.03 |  54000.0000 | 28000.0000 | 2000.0000 | 1022378848 B |        1.49 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |           200 |      10000 |             5 | 14,761,860.197 us |  85,916.1151 us |  80,365.9874 us | 14,771,518.756 us |  1.00 |    0.00 |  99000.0000 | 36000.0000 | 4000.0000 | 1838436032 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |           200 |      10000 |             5 | 15,279,345.513 us |  95,180.9868 us |  89,032.3541 us | 15,314,680.840 us |  1.04 |    0.01 | 114000.0000 | 39000.0000 | 1000.0000 | 2174396520 B |        1.18 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |         10000 |        100 |             0 |  1,122,535.263 us |  22,159.2005 us |  18,503.9360 us |  1,125,777.283 us |  1.00 |    0.00 |  12000.0000 | 12000.0000 | 3000.0000 |  202355400 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |         10000 |        100 |             0 |  1,606,910.709 us |  31,556.2656 us |  30,992.4722 us |  1,602,150.477 us |  1.44 |    0.04 |  21000.0000 | 20000.0000 | 3000.0000 |  368756200 B |        1.82 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |         10000 |        100 |             1 |  2,821,367.698 us |  34,917.3246 us |  30,953.2988 us |  2,818,224.300 us |  1.00 |    0.00 |  21000.0000 | 21000.0000 | 4000.0000 |  346360744 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |         10000 |        100 |             1 |  3,030,322.043 us |  49,233.1532 us |  46,052.7221 us |  3,028,998.425 us |  1.07 |    0.02 |  27000.0000 | 14000.0000 | 1000.0000 |  512757328 B |        1.48 |
|                               |               |            |               |                   |                 |                 |                   |       |         |             |            |           |              |             |
|                AsyncKeyedLock |         10000 |        100 |             5 |  7,612,644.786 us |  79,660.4214 us |  74,514.4076 us |  7,605,484.079 us |  1.00 |    0.00 |  52000.0000 | 21000.0000 | 4000.0000 |  922372528 B |        1.00 |
| AsyncKeyLockFromImageSharpWeb |         10000 |        100 |             5 |  7,660,341.455 us |  46,265.9737 us |  41,013.5806 us |  7,648,538.052 us |  1.01 |    0.01 |  60000.0000 | 23000.0000 | 4000.0000 | 1088784168 B |        1.18 |

Source for benchmarks run on GitHub Actions

Depending on the conditions of the test, SharpLabor.ImageSharp.Web.AsyncKeyLock consumes 18% to 84% MORE memory compared to AsyncKeyedLock, and the time it spends ranges between 1% and 64% more time. Tests with different parameters for number of locks and contention, as well as trying out with a different number of GUID reversals to simulate more realistic environments where something is actually happening inside the lock.

The above benchmarks are from GitHub Actions but I tested on a laptop in Safe Mode and got very similar results.

Code for benchmarks can be found at https://github.com/MarkCiliaVincenti/AsyncKeyedLockBenchmarks/

Would you consider adding a dependency to my library if I make a PR, @JimBobSquarePants?

[MarkCiliaVincenti]

@JimBobSquarePants

[JimBobSquarePants]

Taking a much-needed break over the holiday period.

These benchmarks are definitely very interesting but taking on a new dependency for such a key part of the middleware infrastructure is not something I would do in haste.

It's very likely that the existing implementation can be improved to remove the gap with a little profiling and refactoring so I would attempt that first.

[MarkCiliaVincenti]

The difference in memory use is mainly down to the pooling. However, things can get tricky with regards to pooling and race conditions.

[MarkCiliaVincenti]

I've made some improvements in performance, @JimBobSquarePants.
Please check the results in https://github.com/MarkCiliaVincenti/AsyncKeyedLockBenchmarks/actions/runs/4027111603/jobs/6922445120 out (just ignore the ones for 10000 locks, 10000 contention as those are skipped in reality)

When you say you would not do it in haste it means you're not completely closing the door to this, right?