ZFS Which is better?creating a zpool by a freebsd-zfs partition or a disk?

[Annalesa]

gpart create -s GPT nda0
gpart add -t freebsd-zfs -a 2m nda0
zpool create zpool nda0p1
or
zpool create zpool nda0

 

[vmb]

I use both methods:
On zroot pool for single drives and mirrored pairs I use the partition method as modern machines need an EFI partition.
On data pools with mirrors, RAID and both, I use the whole drive. When a drive fails, swapping in a spare of the same make/model/capacity is a simple process, particularly if it is hot-swappable.

 

[SirDice]

Performance wise it's not going to matter except when the partition is misaligned. For your own sanity however it might be useful to create partitions, with labels, it'll make it easier to identify individual disks.

 

[loveydovey]

There's also that fringe thing that some computers/OSs might try to write stuff to the MBR when you insert a drive without any partition table?

 

[drsnx60]

In the early days of ZFS on SPARC/Solaris ( eg: year 2008 ) we were advised to let ZFS use the whole HDD and not to use partitions. because using partitions would not place the HDD onboard Memory Cache under ZFS control.
Has this issue been solved in OpenZFS 2.x that we are using now. ?

 

[tingo]

In the early days of ZFS on SPARC/Solaris ( eg: year 2008 ) we were advised to let ZFS use the whole HDD and not to use partitions. because using partitions would not place the HDD onboard Memory Cache under ZFS control.
Has this issue been solved in OpenZFS 2.x that we are using now. ?

Yes. This hasn't been an issue on FreeBSD for more than 10 - 15 years now, perhaps even longer.

 

[atax1a]

we strongly recommend against using the whole drive. various things expect there to be partition tables there, and murphy's law says you'll run into surprise data loss when you least expect it.