I find replicating Database through its log files \"as with Postgre\" is a very handy tool, and now with SSDs becoming a performance tool, why not have a NetBSD filesystem journaling more flexible.
that the journal can be logged to a separate drive, where it logs every block of data and meta-data, and using the log structured nature to gain some features we gain with zfs for example, like we replicate the file system through the journal, by for example streaming differences, as with zfs.
i know many will say why we do it while we have zfs, for me actually zfs is not so usable, it consumes much memory, and doesn't fit in virtual environments, and severe fragmentation happen when used with a database, especially if the database engine like most now uses MVCC for concurrency. also i still find UFS with a journal much more reliable, although zfs makes sure to checksum every block, it doesn't pretty much guarantee that these blocks are really written to disk by the time it returned.
so i believe while zfs makes features out of copy on write structure, a continuous log on seperate disk can give us alot of such features while we retain the minimal fragmentation on the filesystem and keep the memory usage to minimum. i am not a kernel developer, but i think since LFS layer exists it can be made use of for such log if that makes any sense.
CVSweb for NetBSD wikisrc <wikimaster@NetBSD.org> software: FreeBSD-CVSweb