NLog is faster now

[applusplus]

I just want to mention, for people who find this Repository, that with the current state NLog is now faster.
At least, if you test it like it was tested in this Repo.
For more details, see my Fork:
https://github.com/applusplus/LoggingPerformance

[snakefoot]

Think it is unfair that you have limited Serilog Async queue to 500 logevents:

https://github.com/applusplus/LoggingPerformance/blob/77272ca44bf1c4df372fbc1fd8718c40a2e21e7d/LoggingPerformance.Serilog/Program.cs#L19

The NLog AsyncTargetWrapper queue is by default 10000 logevents.

[applusplus]

I just adjusted the settings to match NLog configurations made by the author.
https://github.com/applusplus/LoggingPerformance/blob/77272ca44bf1c4df372fbc1fd8718c40a2e21e7d/LoggingPerformance.NLog/Program.cs#L39

EDIT: Also I want to add, that in my opinion this project is not enough to make a proper performance comparison. But it's one of google's first results, if you look for a benchmark.

[snakefoot]

NLog AsyncTargetWrapper BatchSize is not the size of the in-memory-queue, but how many items it will write in a single batch, when the background-thread is eating from the queue:

https://nlog-project.org/documentation/v4.7.0/html/P_NLog_Targets_Wrappers_AsyncTargetWrapper_BatchSize.htm

Serilog Async BufferSize should match NLog AsyncTargetWrapper QueueLimit (Both have default 10000):

https://nlog-project.org/documentation/v4.7.0/html/P_NLog_Targets_Wrappers_AsyncTargetWrapper_QueueLimit.htm

[applusplus]

You are right. I removed the buffer limit, and it's now even slower... 🤔

[snakefoot]

Sounds like the Serilog Async Sink is punished by using BlockingCollection for its in-memory-queue.

But the idea is that the application logger-threads should produce into the in-memory-queue without getting blocked, so the queue should have a reasonable size.

[snakefoot]

Created #6 with your updated benchmark-results.

[demigor]

Removed LibLog, upgraded to latest packages (NLog 5.2.8, Serilog 3.1.1), .NET 8 and called loggers directly.

My results (Dell XPS Desktop, Windows 11):

Serilog
Stopwatch.IsHighResolution = True

Generated 5000000 values in 3382ms (throughput = 1478167.889 ops per second)
Mean latency = 0.656mcs
99% observations less than = 3.500mcs
99.99% observations less than = 67.600mcs
0) 0.300 - 0.6200 = 4134983
1) 0.620 - 0.9400 = 233408
2) 0.940 - 1.2600 = 193237
3) 1.260 - 1.5800 = 94794
4) 1.580 - 1.9000 = 210698
5) 1.900 - 2.2200 = 31888
6) 2.220 - 2.5400 = 10044
7) 2.540 - 2.8600 = 2754
8) 2.860 - 3.1800 = 5841
9) 3.180 - 3.5000 = 38146
Press any key to stop...

NLog
Stopwatch.IsHighResolution = True

Generated 5000000 values in 1621ms (throughput = 3084046.750 ops per second)
Mean latency = 0.306mcs
99% observations less than = 0.600mcs
99.99% observations less than = 259.100mcs
0) 0.000 - 0.0600 = 1740563
1) 0.060 - 0.1200 = 2386799
2) 0.120 - 0.1800 = 0
3) 0.180 - 0.2400 = 531129
4) 0.240 - 0.3000 = 176113
5) 0.300 - 0.3600 = 176113
6) 0.360 - 0.4200 = 69041
7) 0.420 - 0.4800 = 0
8) 0.480 - 0.5400 = 36624
9) 0.540 - 0.6000 = 21122
Press any key to stop...

NLog is 2x faster than Serilog.