Re: installing amd64 kernel
On Thu, Dec 22, 2005 at 08:37:47PM +0000, Adam James wrote:
> Could you explain some of the reasoning behind your comment?
>
> All the partitions on my Linux systems are formatted with reiserfs
> (300GB+ of actual data), and I've yet to experience any problems.
>
> Making a statement inferring that there are well known bugs in a widely
> used piece of software without explaining what they are doesn't really
> help anyone make an informed decision.
Well I used reiserfs 3.5 a bit and then used reiserfs 3.6 for probably 2
years before finally going to ext3. I had files corrupted when systems
lost power in the middle of a write ending up with partially new
contensts, and partially old contents and often partially old deleted
data. ext3 in the default mode of journal=ordered doesn't ever do that.
I either get the old file or the new file. I have never gotten
something in between or worse, something else random.
Using reiserfs on a laptop without proper suspend support for auto
suspending when the battery runs low makes you very likely to encounter
this type of corruption. Sure the fsck isn't necesary because the meta
data is perfect. The file content isn't though. ext3 only updates the
meta data after the file content is written to disk so you avoid random
data when adding to a file. I believe (but could be wrong) that ext3
also writes changes to unused blocks, then updates meta data and frees
the old blocks.
If reiserfs 4 does full journalling (which I believe it does) it will be
a major improvement. It is still a very complex filesystem, and
personally I prefer something that you can actually fix if it breaks
using some kind of fsck. At least xfs seems to have a decent repair
tool, while reiserfs still doesn't seem to. The rebuilddb option often
does more harm than good from the experiences I have had as well as
other users I have read messages from on reiserfs. I just want a
filesystem that stores my files, doesn't make a mess of the contents,
and can be fixed if it breaks without major data loss. I don't care if
it can delete 100000 files per second. I don't really ever need to do
that. I need to sometimes write files or change files, and most of the
time read files. Reasonably fast dir lookups are nice, but ext3 has
that too. Most things that do lots of I/O use pre created files and
hence don't care how fast meta data updates can be done, while raw file
i/o is limited almost entirely by disk speed.
The 2.6 bugs in XFS seem to be known, although last I checked they still
hadn't been tracked down.
So that is why I only use ext3 now, and only recomend ext3.
Len Sorensen
Reply to: