On Mon, Mar 20, 2023 at 07:40:44PM -0400, Martin K. Petersen wrote: > > Luis, > > > /sys/block/<disk>/queue/minimum_io_size > > > > The documentation suggests " For disk drives this is often the > > physical block size." > > > > /sys/block/<disk>/queue/optimal_io_size > > > > The documentation suggests this is "This is rarely reported for disk > > drives." > > min_io and opt_io are used to key mkfs.xfs' sunit/swidth. So if you're > using a hardware RAID, MD, or DM, we'll attempt to align allocations on > stripe boundaries. > > Back when that "rarely reported" blurb was written (2009), we did not > have any individual disk drives which reported min_io/opt_io. Reporting > those parameters was mostly a storage array thing. These days it's > fairly common for both disk drives and SSDs to fill out these fields. Glad you mentioned this, I followed up in my review of these and I see even the names, swidth, sunit, are all "stripe" indicative. Based on what you are saying, it seems we may need to update docs to reflect actual / new uses. > > From my review of xfs's mkfs is we essentially use the physical block > > size as a default sector size if set, otherwise we use the device's > > logical block size if set otherwise xfsprog's default and so 4096. > > Yep. The use case for swidth / sunit on mkfs.xfs seemed pretty tied to striping, and it was not obvious or clear / if using it for hints could be used today as perhaps intended. At least all the naming and validation stuff seems to make it very "stripy" still. Luis
