XFS

Unlike ext3, XFS was designed by SGI for efficiently journaling from its beginning, rather than having journaling added on to an existing filesystem. As you might predict, the result is a bit faster than ext3, some of which is just from better efficiency in the journal implementation. However, part of this speed results from the fact that XFS only journals metadata, and it doesn't even have an option to try and order the data vs. metadata writes. Accordingly, XFS is most like ext3's writeback mode. One critical difference is that in situations where garbage blocks may have been written to a file, the main concern with ext3 writeback, the journal playback in XFS will instead zero out these entries. Then they are unlikely to be interpreted as real data by an application. This is sufficient to keep PostgreSQL from being confused if it tries to read them, as a zeroed block won't have the right header to look like either a WAL block or a database data block. Some consider...