Improved Hypervel vs Octane benchmarks

[binaryfire]

Check List

• Provide a clear title and detailed description of your question.
• I have searched the existing discussions and documentation to see if my question has already been answered.
• Include any relevant code snippets or examples to help us understand your question better.
• We encourage you to help others in the community by answering questions or providing feedback on existing discussions. This helps us build a strong and supportive community around Hypervel.
• Please be respectful and considerate in your interactions with others in the community.
• Any questions labeled not enough info will be ignored and closed after 3 days.

Content

I've had several questions regarding the validity of the Hypervel vs Octane benchmarks when speaking to Laravel devs, because they use a different number of threads for Octane (4) and Hyperf (16):

Is it possible to redo the benchmarks with the same number of workers for each? This would make them much more convincing. Also, putting the testing code in a repo so people can reproduce them would be good.

[albertcht]

Hi @binaryfire, the thread count here is for the wrk tool. The number of workers for Octane and Hypervel is the same (8 workers).

In our benchmark testing for Hypervel, we set the wrk thread number to 16 because during the one-second simulated I/O request tests, the wrk tool itself encounters performance bottlenecks. Therefore, we need to increase wrk's thread count to enable it to send more requests. In the case of Octane, regardless of how many threads wrk uses, the test results will always be limited by the number of workers that Octane itself starts.

[binaryfire]

Ah ok, got it. I just saw “worker numbers are configured as CPU cores by default”.

It’d be good to make this text larger. Or even better, put “Octane (8 workers)” and “Hypervel (8 workers)” above each set of results. I think it will help avoid confusion.

[albertcht]

good idea! I've updated the docs.