Change Details

Quite surprisingly to me, it doesn't seem to immediately panic. So I'm probably missing something. ;-)

Quite surprisingly to me, it doesn't seem to immediately panic. So I'm probably missing something. ;-) In our getblk-heavy workloads (which have other scaling limitations beyond those in FreeBSD), on 2x48 CPU (2P) metal, we see significantly better scaling with this change at 6+ threads (4.3x single-thread vs 3.7x single-thread), and the gap gets wider at higher thread counts. I don't have any great benchmarks on vanilla FreeBSD.

Quite surprisingly to me, it doesn't seem to immediately panic. So I'm probably missing something. ;-) In our getblk-heavy workloads (which have other scaling limitations beyond those in FreeBSD), on 2x48 CPU (2P) metal, we see significantly better scaling with this change at 6+ threads (4.3x single-thread vs 3.7x single-thread), and the gap gets wider at higher thread counts. I don't have any great benchmarks on vanilla FreeBSD.