Experiences, comments -- largely positive.
General comments:
- Installation required snarfing the patch and applying it to the kernel. No issues there.
- There's a little dance you have to go through if you're converting an existing GNU/Linux installation to reiserfs -- which you're very likely doing as few GNU/Linux installs (I know of none) roll out reiserfs by default. It's called the reiserfs shuffle, and involves backing off your data, reformatting the partition, and restoring the data. For all partitions but / (root), this can be done without rebooting, preferably from single-user mode. For your root partition, you'll have to boot an alternative system (more below). My own suggestion is to create two root filesystems, one to be used as a backup and/or for maintenance, which greatly simplifies any number of issues.
- Very small partitions and reiserfs don't go together awfully well. There's a certain amount of overhead (forget what) that's required. Anything below about 10-20MB is going to suffer significant loss to journal overhead. You're also not saving much in a reiserfs core benefit: fast crash recovery. An fsck of a 10MB partition runs in a matter of seconds, about the same as a journal rollback. What might such small partitions be? Typically, /boot, possibly /tmp. There's the additional issue that LILO and reiserfs don't interact too well (though grub's supposed to be better). I've created an ext2fs /boot partition and mount it read-only, except when performing system admin.
- There were issues with reiserfs corruption in some early 2.4.x series kernels. These seem to have been resolved as of 2.4.4 or thereabouts -- with current kernels (2.4.8 was released today) there's little issue. The 2.2.x series was never effected by these issues AFAIK.
- My installation is on a laptop, as noted. The box is known to be shut down suddenly, run out of battery power, and, it went through a period of crashes (bad power connection compounded by poor Speedstep CPU support in GNU/Linux). Never suffered any corruption, watched my journals replay in 1-3 seconds, for partitions of up to 10 GB. Very nice.
- Documentation for reiserfs still lags. Responded to a recent debian-user issue in which ext2fs mount options (errors=remount) turned out not to be supported under reiserfs. The current Debian mount (1) manpage doesn't specify reiserfs options.
- As noted above, you may need a rescue disk for your reiserfs system. Problem is that most popular rescue systems don't have reiserfs support, including Tom's Root/Boot. One exception is recent versions of the LinuxCare BBC (bootable business card), version 1.5 of which includes reiserfs support. Other tools can be modified to provide such support, so explore your options.
- Though reiserfs was the first of the generally available GNU/Linux journalling filesystems, there are now several to choose from, including ext3fs, XFS, and JFS. Word is that there are utilities to convert directly from ext2fs filesystems to ext3fs. Various performance claims are also made among the various filesystems.
- Performance: mostly anecdotal. Some people argue that journaling is inherently slower than a write-behind cached filesystem (e.g.: ext2fs). OTOH, there was an interesting post, to LinuxToday, IIRC, in the past few weeks arguing that for certain types of (relatively common) metadata transactions, journaling can in fact be much faster. Naturally, there are the usual partisan debates over which journaling FS is fastest.
Cheers.
