On Sat, Mar 30, 2024 at 12:40 PM Tomas Vondra <tomas.von...@enterprisedb.com> wrote: > Sorry, I meant the prefetch (readahead) built into ZFS. I may be wrong > but I don't think the regular RA (in linux kernel) works for ZFS, right?
Right, it separate page cache ("ARC") and prefetch settings: https://openzfs.github.io/openzfs-docs/Performance%20and%20Tuning/Module%20Parameters.html That's probably why Linux posix_fadvise didn't affect it, well that and the fact, at a wild guess, that Solaris didn't have that system call... > I was wondering if we could use this (posix_fadvise) to improve that, > essentially by issuing fadvise even for sequential patterns. But now > that I think about that, if posix_fadvise works since 2.2, maybe RA > works too now?) It should work fine. I am planning to look into this a bit some day soon -- I think there may be some interesting interactions between systems with big pages/records like ZFS/BTRFS/... and io_combine_limit that might offer interesting optimisation tweak opportunities, but first things first...