Recommendations for running hyperscan on multicore setup

[venkatsvpr]

Hello,

This is not directly a issue on the hyperscan package. Posted it here if incase folks have some suggestions.

I'm currently experimenting with using Hyperscan for regex matching in a test service written in Go. To interact with the Hyperscan library, I'm using the flier/gohs package.

So far, I've had a great experience using Hyperscan on a single-core machine. However, when I tried running it on a multi-core machine, I didn't observe any significant performance improvements. You can find more details about this issue in this link: https://github.com/flier/gohs/issues/56.

The tests in the link are benchmark tests written in Go. I do see from the blog that the performance increases linearly with the cores. https://www.intel.com/content/www/us/en/developer/articles/technical/introduction-to-hyperscan.html).

I am unable to understand why the perf doesn't increases linearly. I'm not sure if I'm missing any configuration or if I'm testing it correctly.

Do you have any suggestions?

Thank you!

[variar]

I use hyperscan from c++ application. To utilize several cores I split the input between them, and then merge the results. This way I do see almost linear speedup.

[hongyang7]

@venkatsvpr Hyperscan is a pure single-core matching library, not involving any multi-thread design. The scalability on multi-core generally means when facing multiple inputs and one pattern set, different core is for different scanning thread on different input, but with shared copy of the same bytecode (read-only).