Ext3/Ext4 data integrity guarantees
Posted May 29, 2008 20:52 UTC (Thu) by anton
Parent article: Quotes of the week
I turn off IDE disk write caching on all machines I maintain in order
to really get the data integrity that journaling promises, and Andrew
Morton drops data integrity patches because he worries about the speed
impact of barriers and flushes (which are much less). If Linux used
barriers and flushes to get data integrity, and I knew about that, I
could turn on the write cache and see a speedup. However, I did not
come across information about such issues despite being interested,
and looking around now and then (and what I read did not inspire
confidence), so I went ahead and disabled the write cache.
If I cared more about speed than data integrity, I would use XFS
instead of ext3 in the first place. OTOH, if the Linux maintainers
don't care for data integrity, maybe I should be looking for another
kernel. Fortunately, looking through the thread, it seems that sanity
seems to have come back to the Linux maintainers.
Concerning what can be lost with write caching enabled and without
barriers, I have run a little
test several years ago on two disk drives, and they both lost data
that was several seconds old(er than data that ended up on the disk).
to post comments)