[OpenIndiana-discuss] HDD Upgrade problem

[Jim Klimov]

2012-06-12 11:24, michelle wrote:
> Is there a way for the system to me a bit more clever; we have 4k
> sectors now but how long before that is upped again; maybe another decade?
>
> After all, if it is a redundant set, what is the reason why it can't
> just dump the disk to be replaced and rebuild on the new drive?
>
> I know I'm being a noob here as I don't have the in depth knowledge of
> many people on this list ... but from my "user" perspective, it strikes
> me that this is the more obvious question.

You can look in the zfs-discuss list archives of the past month,
I asked about how scrubbing and resilvers work, dumping my
understanding onto the list and asking for corrections :)

Basically, a redundant/striped ZFS data block is striped over
several sectors of hard disks, with chunk sizes based on the
ashift value. Rebuilding the same pool onto a new disk requires
that these chunks have the same size - even if they no longer
match (are smaller than) native sector size, leading to various
possible problems regarding performance and reliability.

On the other hand, chunks larger than the typical metadata
block size (of which there are plenty in ZFS), even these 4KB
minimal writes, bump the pool's storage overheads considerably.
Many people have noticed the inflation of consumed disk space
when migrating their data from an old pool onto a new one.
But this is unnoticeable on Windows and other dominant OSes
which use 4KB cluster/block sizes by default for many years
now, and HW RAID controllers often use yet larger "chunks"
ranging 16-64KB to lay out userdata onto RAIDed disks.

That's the short story :)
//Jim Klimov