Quote: “ESXi, which already has the features of ZFS and more”

ESXi does not have the data integrity checks and consistency guarantees of ZFS.

Quote: “If I were bare metal, then I would use ZFS for its self-healing and robustness”

Running it at a lower level is certainly preferable, but ZFS can still detect and report consistency problems even when running in disk images, or over multiple layers. Colin Percival, the maintainer of the Amazon EC2 FreeBSD images, is working on a ZFSonRoot version, and OrionVM FreeBSD images use it by default.

Quote: “and easier and more solid to administer because it works even if the operating system will not boot”

ZFS is the easiest file system to administer because it handles volume management, partitioning, formatting, integrity checking, RAID-like features for multiple drives, and inline encryption in the latest OpenZFS versions. If the OS won’t boot, you can boot a helper VM or FreeBSD installer, drop to a shell, and import your ZFS pools. I can’t quantify “solidness”, but it’s the only file system I trust.

Quote: “ZFS is also hard on resources on a heavily loaded server.”

In the words of Allan Jude, one of the principle ZFS architects at FreeBSD and the author of several of the authoritative ZFS books: ZFS will use the resources you give it. Even ZFS in a constrained environment (low or limited resources, no ECC memory, etc) is a more robust and trustworthy system than anything else today.

Quote: “ZFS is mainly used to deploy filesystems on bare metal machines with multiple attached storage uints or HDDs.”

ZFS hasn’t been limited to this use case on FreeBSD for years. In particular:

* The official Poudriere system for building ports in the new pkgng system uses ZFS to create isolated pools that are blown away once a port is built. Whenever you install a package on FreeBSD now, it’s been built this way. It’s also how you build custom ports now.

* It’s invaluable for laptops or consumer-grade desktops with flaky hardware, because it will always remain consistent even if a suspend/hibernation fails, or a battery dies. You can’t say the same for UFS (or ext4, or xfs, etc), even with journaling or softupdates enabled.

* iocage uses ZFS by default for creating jails, and it’s increasingly the preferred approach for ezjail and manual builds. Every other approach for using jails is at the very least tedious (I would know, unfortunately!)

* Allan Jude demonstrated at AsiaBSDCon 2019 the idea of using ZFS for in-place, atomic upgrades of the base system. This is the use case I’m most excited about, and would bring upgrades for FreeBSD closer to what Solaris has (or had, rest in peace).

Cheers

Yes, IT_Architect has correctly spotted the right answer. ZFS is mainly used to deploy filesystems on bare metal machines with multiple attached storage uints or HDDs. In virtual machines ther’s no point in using ZFS, because the underlying hard disk is nothing than a single file created on a already existing storage and filesystem.

I don’t know about him, but I only use UFS because I install on ESXi, which already has the features of ZFS and more, and easier and more solid to administer because it works even if the operating system will not boot. If I were bare metal, then I would use ZFS for its self-healing and robustness, but ZFS is also hard on resources on a heavily loaded server.