3 ms·

The thing I miss most in ZFS (aside from a native in-kernel linux port, as I've found ZoL to be kinda buggy) is to be able to grow zdevs. I can add a new zdev,

The thing I miss most in ZFS (aside from a native in-kernel linux port, as I've found ZoL to be kinda buggy) is to be able to grow zdevs. I can add a new zdev, I can remove them, but I can't add one disk to my 7-disk RAIDZ2. MD RAID has no problem with this. Also, if you add zdevs later, you will never get full performance again as the striping will be off (putting more writes on the zdevs with more free space), and there's no 'rebalance' kind of thing (I've seen some scripts that just move stuff around, but I've never seen any evidence that they work, and I really doubt that they do based on my understanding of zdev allocation).

nathan7 11y ago
I'm assuming you mean vdevs — you can resize zvols (virtual block devices) just fine. (and you can even grow disk-type vdevs, though they'll be somewhat unbalanced)
- throwaway7767 11y ago
  Yeah, sorry, edited. It's mainly the ability to resize RAIDZ or RAIDZ2 zdevs that would be really helpful.
Scramblejams 11y ago
Me too. I've added so many drives in onesies and twosies to so many MD RAID5/6 arrays over the years that I simply can't confidently deploy ZFS for that case.
cmurf 11y ago
md RAID will require a lengthy reshape. BTRFS supports adding a new drive, reshape (balance) is optional.
- throwaway7767 11y ago
  > md RAID will require a lengthy reshape Array reshapes are rare, and it'll only use idle IO bandwidth anyway. Redundancy is guaranteed during the whole process. So I don't see this as a problem. > BTRFS supports adding a new drive, reshape (balance) is optional Thanks for pointing this out. I didn't know btrfs supported array reshapes. I'm waiting for it to mature (RAID5/6 was only added in 3.19, for example) but I expect I'll switch over at some point, probably at the point it's considered "boring" (exactly the experience I want from my filesystems).

nathan7 11y ago

I'm assuming you mean vdevs — you can resize zvols (virtual block devices) just fine. (and you can even grow disk-type vdevs, though they'll be somewhat unbalanced)

throwaway7767 11y ago
Yeah, sorry, edited. It's mainly the ability to resize RAIDZ or RAIDZ2 zdevs that would be really helpful.

Scramblejams 11y ago

Me too. I've added so many drives in onesies and twosies to so many MD RAID5/6 arrays over the years that I simply can't confidently deploy ZFS for that case.

cmurf 11y ago

md RAID will require a lengthy reshape. BTRFS supports adding a new drive, reshape (balance) is optional.

throwaway7767 11y ago
> md RAID will require a lengthy reshape Array reshapes are rare, and it'll only use idle IO bandwidth anyway. Redundancy is guaranteed during the whole process. So I don't see this as a problem. > BTRFS supports adding a new drive, reshape (balance) is optional Thanks for pointing this out. I didn't know btrfs supported array reshapes. I'm waiting for it to mature (RAID5/6 was only added in 3.19, for example) but I expect I'll switch over at some point, probably at the point it's considered "boring" (exactly the experience I want from my filesystems).