Rob Owens wrote:
The resilience is due to the way the journal is written, if I
understand correctly. Maybe somebody on this list who understands it
better can confirm or deny. There is a journal_data_writeback option
for ext3 which will speed up writes to the filesystem, but reduce its
resilience to power loss. With this option enabled, I recall reading
that the ext3 benchmarks are pretty similar to XFS.
Yep. As always, LWN probably has the best word on it [1].
Short answer: ext3 is outdated, ext4 is current and can still be configured
to get the same "better data resilience" without losing all its benefits.
XFS should also be able to do so. Criticising ext4 for data resilience
"problems" and praising XFS is a fallacy, both go in the same direction.
Now the debate is around the default configuration of modern filesystems
(basically performance vs safety). As YMMVVM (very much), one should
probably just ignore the debate, take 30m to learn about the issue, and
configure his filesystem properly.
Well, opinions. ;-)
For stable users using ext3, writeback can theoretically offer better
throughput, as it doesn't force data to be be pushed on the platters before
the metadata has been committed to the journal. It still keeps the
filesystem consistent (the only thing a journal is supposed to do), but the
risk of corrupting the data is greater. I, personally, don't seek to
minimize that risk, I want it to be zero -- no filesystem can help here, and
no filesystem will ever do. That's one reason why I don't like to see ext3
recommended for its data resilience: it gives the user the illusion of safety.
Of course, it still makes sense to minimize the risk in certain scenarios
where it can't be eliminated; but again, modern filesystems can be
configured to do so.
-thib
[1] http://lwn.net/Articles/322823/
--
To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/4bd9d34c.1010...@stammed.net
