[pve-devel] ZFS alternatives

Andreas Steinel a.steinel at gmail.com
Sun Feb 25 21:30:42 CET 2018


On Sun, Feb 25, 2018 at 9:08 PM, Gandalf Corvotempesta
<gandalf.corvotempesta at gmail.com> wrote:
> As I can see, old data still keep the old parity, with MD all data are
> rewritten thus making use of all disks for performance reason

Yes, data has to be "resynched" manually via send/receive - same
way as with an expansion of disks.



> AFAIK the only real reason to use ZFS is for bitrot and incremental backups
> using shapshots.

CoW clones, variable block size, transparent compression, deduplication,
resync only used records on disk failure


> Everything else can also be made by using MD plus some additional software
> (in pure Unix style: one software to do one thing. ZFS it's the opposite
> :one software for all: raid, filesystem, snapshots, volume manager,
> backups,....)

Yes, ZFS kinda breakes this concept, yet I see the charme in it. It happend
 - more than once - with e.g. LVM that you have enough space, yet not all in
one volume, but in multiple. This cannot happen in ZFS because of the unified
approach that all data is stored in the pool and you can use the free space
everywhere. I really think combining this is the way to go.



More information about the pve-devel mailing list