indeed, pretty incredible speed ups.
Is there a process, that FreeBSD release goes through, to confirm that the performance gains did not bypass, accidently, any of the previously enforced security and isolation features ?
Simply put, the FreeBSD 13-BETA1 performance we are seeing so far on Intel x86_64 hardware is nothing short of terrific albeit overdue – this isn’t a magical performance boost but likely due to not fully leveraging the hardware previously, which would be the case especially if these gains are chalked up due to P-State / power management.
E.g., there was an issue where the hwpstate would get stuck in the slowest possible P-State:
There’s some talk of regression testing in the official documentation, and I’ve read somewhere sometime that FreeBSD runs regular benchmarks to see that things are going in the right direction performance wise.
It’s incredible that kernels as old as FreeBSD’s can still have such relatively low hanging fruits. One would’ve thought that only incremental improvements would be possible at this stage.
As mentioned elsewhere, the big perf gap in compute-intensive benchmarks suggests that the phoronix benchmark environment was having issues with CPU power states before 13.
In terms of scalability though, there are indeed lots of incremental improvements (in the VFS/cache area especially) but they do eventually add up.
FreeBSD 12.2-RELEASE sounds like a sudden surprise. I believe there’s a typo in the title.