[NCLUG] ext3 versus ReiserFS?

[John L. Bass]

Mike Loseke writes:
	 I always find it interesting when people evangelize different JFS setups,
	their number one remark is fast fsck/mount times. This is nice or course
	but has always seemed misplaced to me.

Generally JFS's have the same fsck time, they just claim not to need it as often
by being able to replay the journal for mounts following a crash. There are a
number of other ways that filesystem meta data can get scrambled, including the
journal during power fail crashes. Having filesystems that have a robust meta
data ONLY minimizes filesystem corruption due to inode, link ptr, and freelists.

In general JFS's offer no protection against data corruption, and sometimes no
protection against directory corruption.

Also JFS's assume that there is no write buffering or write reordering in the
IO stream - things like RAID devices can break this assumption and leave the
journal and filesystem meta data trashed in certain types of crashes.

Building fault resisitance into a facility takes a lot more design than tossing
a UPS and JFS at the systems.

John