On 19 Apr 2002, Eric Sandeen wrote: > re: the "nulls in files" problem, it's a fact of life that on a metadata > journaling filesytem, if you lose power, you lose data. In the case of > xfs, synchronous transactions used to mean that you sometimes wound up > with files with a length, but no data if you crashed before a sync. > Most of these transactions have now been made asynchronous, and you're > much more likely to just get your "old" data back after a crash, since > metadata is not forced out before file data. Is this now equivalent to ext3's default "ordered data" mode then? It sounds like this would make xfs more reliable than reiserfs for me.
