To quote his “I would recommend against per-thread caches. Instead, make the per-cpu caches actually *be* per-cpu (that is, not require a mutex).”

Then, he has pointed out that there is potential problems in FreeBSD’s SLAB allocator which is different from DragonFly’s. (Not too surprisingly, he is again advertising his IPI message approach).

I am thinking about this seriously. With a per-CPU approach, we will reduce cache misses drastically when there is no heavy CPU migrations.