[Discuss] Thin Provisioned LVM

[hag at linnaean.org (Daniel Hagerty)]

markw at mohawksoft.com writes:

> Again, like I said, these do not address the problems. Specifically, the
> post about sparse volumes says nothing about how to keep a ZFS pool from
> growing out of control on a sparse presented to it from a SAN. It merely
> says give ZFS whole disks, which is stupid.

    What you are looking for is ATA TRIM support.  It is an ANSI T.13
standard, which should tell you that your ire with ZFS is misplaced.

    *If*, and only *if* your disk abstraction supports ATA TRIM, it is
possible for filesystems to communicate with the disk abstraction about
blocks they have previously allocated, but are now free.  ZFS has the
problem, ext* has the problem, and in fact, anything that allocates
blocks on a disk abstraction has the problem.  As you yourself implied,
for many decades, leaving garbage blocks didn't matter and so there was
no way to communicate this to the disk.

    Your operating system may or may not support all the needed parts,
because all of this is still bloody.  Needing to free blocks from the
spinning rust emulation is new.  The OS, ZFS, and disks I'm running can
all do TRIM, but I can't speak for yours.