On Sat, Dec 15, 2007 at 07:12:18PM -0800, Tracy R Reed wrote:
>
> Soft-updates was just something BSD did when they didn't have a journaled 
> fs. Overall I would much rather have full journaling than just 
> soft-updates.
>

You've glossed over the major win for softupdates. It feels
fast.  Substantially. True, the actual IO is about the same as
UFS without softupdates, but the vitalization that's safe with
atomic writes results in comand lines that return immediately vs
150 or 1500 ms. And that's realized more than IO. It may not seem
like much, but users notice enough to say it's faster.  It may
help for busy systems with a lot of writes, not sure.

journaling, hardware/software raid, battery, reiser, ext2, ext3,
etc is a whole nother discusson. performance, cost, administrative
requirements are different for everyone. some people are using LFS
(logging filesystem, journaled) on NetBSD but I don't think it's
past the bugs/time sweet spot yet.

I've heard of some pretty nasty file corruption on "clean" reiser
partitions, but there is risk everywhere, just has different flavors.

// George

-- 
George Georgalis, information system scientist <IXOYE><


-- 
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list

Reply via email to