Deprecating ReiserFS

ReiserFS is getting deprecated from Linux, mostly due to it not being ready for the year 2038. This is a little sad, as I still use it on some systems for storing the Gentoo Portage tree, and the Linux kernel sources. It works well for this because it supports tail packing, a form of block suballocation, which can save disk space.

So, what alternatives are there for ReiserFS? After asking around and reading some comments around the Internet, I’ve narrowed it down to 3 potential candidates, bcachefs, btrfs, and zfs. Each comes with their own pros and cons, as things tend to do.

bcachefs

There are several downsides for bcachefs for me. The first one I found was that the documentation on their main site seems a bit lacking, followed shortly by finding that there are no ebuilds for it in Gentoo.

Since it was suggested several times on comments on a certain orange site, I asked around if it at least supported block suballocation, which is the main reason I would want to use it anyway. The answer came back as a “no”, so I could safely ignore it for the rest of the journey.

BTRFS

BTRFS seems like a more serious contender. It supports block suballocation, and has good enough documentation. As an additional benefit, it is supported in the mainline Linux kernel, making it easy to use on any modern setup. There are a few issues, such as having to rebalance in certain situations, and this rebalancing can itself cause issues. The files I’m storing are relatively easily recreated with a single git clone, or downloading a tarball and unpacking that, so that doesn’t have to be problematic to me.

ZFS

The final contestant, ZFS, supports block suballocation and has great documentation. It is not part of the mainline Linux kernel, however, so this may make things more complex on some systems. I run ZFS already on a few machines, but not all, so where it is not used already, it is a drawback.

Since my main concern is storing many small files, I created a few logical volumes (and 1 ZFS subvol) and cloned the main reason for wanting a filesystem with block suballocation, the Gentoo Portage tree. The cloning itself was done with --depth=1.For reference, I also created an ext4 volume.

/dev/mapper/edephas0-test.btrfs     5.0G  559M  3.8G  13% /tmp/test/btrfs
/dev/mapper/edephas0-test.ext4      4.9G  756M  3.9G  17% /tmp/test/ext4
/dev/mapper/edephas0-test.reiserfs  5.0G  365M  4.7G   8% /tmp/test/reiserfs
tyilstore0/test                     5.0G  1.1G  4.0G  21% /tmp/test/zfs

Looking at the output from df -h, ReiserFS seem to still be a clear winner when it comes to storing many small files. Nothing is even close. What does surprise me, however, is that ZFS is actually resulting in the largest space requirement. I’m not sure why this is, as it should support block suballocation just fine according to the filesystem comparison chart on Wikipedia.

BTRFS comes out as the next best option, after ReiserFS, so that’ll be what I am going to use on my systems for storing large trees of small files.