2012-03-07 17:21, Edward Ned Harvey wrote:
From: zfs-discuss-boun...@opensolaris.org [mailto:zfs-discuss-
boun...@opensolaris.org] On Behalf Of luis Johnstone
As far as I can tell, the Hitachi Deskstar 7K3000 (HDS723030ALA640) uses
sectors and so I presume does not suffer from such issues (because it
doesn't lie about the physical layout of sectors on-platter)
I think what you mean to ask is "Is the HD7K3000 a piece of junk?" Because
any disk which is lying about its physical sectors is a piece of junk,
regardless of what filesystem is going to be on it.
This isn't a ZFS question. (Nothing wrong with asking - I'm not trying to
discourage having the discussion, but please don't associate such problems
with ZFS as if ZFS is unique in that way.)
Well, of the currently-used FSes, ZFS does have certain issues
with 4k drives, which make it rather unique.
Table/bitmap based FSes like NTFS or FAT can preallocate their
file-allocation tables and don't have any metadata overhead
compared to 512b-sectored disks. In part this is due to having
4kb clusters as default for quite a while, so no expectations
of users change.
ZFS tree has many small nodes as well as file "tails", so it
can effectively utilize small 512b blocks on disks. When you
migrate same pool to ashift=12, it explodes to require more
disk space (from several percent up to ten-twenty, according
to internet rumours). Users don't often expect that, so it is
sorts of an issue.
While it can be speculated that other FSes already steal this
slack space from users on any other drives, this change of
behavior on ZFS on different drive types can be seen as a
drawback, by some.
zfs-discuss mailing list