On Tue, Mar 12, 2024 at 06:48:36PM -0700, Darrick J. Wong wrote: > Hi all, > > For a long time, the maintainers have had a gut feeling that we could > optimize performance of XFS filesystems on non-mechanical storage by > scaling the number of allocation groups to be a multiple of the CPU > count. > > With modern ~2022 hardware, it is common for systems to have more than > four CPU cores and non-striped SSDs ranging in size from 256GB to 4TB. > The default mkfs geometry still defaults to 4 AGs regardless of core > count, which was settled on in the age of spinning rust. > > This patchset adds a different computation for AG count and log size > that is based entirely on a desired level of concurrency. If we detect > storage that is non-rotational (or the sysadmin provides a CLI option), > then we will try to match the AG count to the CPU count to minimize AGF > contention and make the log large enough to minimize grant head > contention. > > If you're going to start using this code, I strongly recommend pulling > from my git trees, which are linked below. > > This has been running on the djcloud for months with no problems. Enjoy! > Comments and questions are, as always, welcome. > > --D Series... Reviewed-by: Bill O'Donnell <bodonnel@xxxxxxxxxx> > > xfsprogs git tree: > https://git.kernel.org/cgit/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=mkfs-scale-geo-on-ssds > > fstests git tree: > https://git.kernel.org/cgit/linux/kernel/git/djwong/xfstests-dev.git/log/?h=mkfs-scale-geo-on-ssds > --- > Commits in this patchset: > * mkfs: allow sizing allocation groups for concurrency > * mkfs: allow sizing internal logs for concurrency > --- > man/man8/mkfs.xfs.8.in | 46 +++++++++ > mkfs/xfs_mkfs.c | 251 +++++++++++++++++++++++++++++++++++++++++++++++- > 2 files changed, 291 insertions(+), 6 deletions(-) > >
