8 comments

[ 3.3 ms ] story [ 31.8 ms ] thread
Wonder why FreeBSD is that slow at ZSTD compression tests ...
I'd be amazed if it isn't a configuration error of some kind.
I'm guessing much of the OpenBSD trailing performance on many benchmarks is due to both memory allocation and filesystem performance. malloc has a lot of security features out-of-the-box; turning them off (don't do that at home) might improve the speed.

But basically these are not benchmarks but conversation topics. A "real" benchmark runs your own code, not someone else's.

As an interesting topic, the LAME result went from 9.94 seconds (OpenBSD 6.4, 2018, Intel Core i9-7980XE) to 18.476 seconds (OpenBSD 7.0, 2021, Intel Core i9 10980XE). So many things changed over those 7 releases....

I would also guess that OpenBSD's SMP implementation is userland mostly. When it comes to kernel its still one big GIANT LOCKED so that is also one of the reasons why the performance is not there.
Actually, that guess is incorrect, it's got a native pthreads with SMP support. Just not great support (for example, OpenMP is not available).

There is ongoing work to avoid the big lock and much of the network stack, for instance, is unlocked.