Page fault performance
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Fri, 05 Apr 2024 16:00:34 UTC
Hello, I was testing FreeBSD-13.2 page faulting performance on AMD64, where a thread allocates a large segment of anonymous memory via mmap(2) and then modifies every 4KiB page there. Transparent large pages were deliberately disabled via vm.pmap.pg_ps_enabled flag. I only want to test the base 4KiB page fault performance. There is good scalability up to 4 concurrent threads, but after that performance follows a somewhat flat line. Any idea if there is some runtime setting to improve concurrency and decrease lock contention in the virtual memory subsystem? Thanks.