Bryan-TheBS-Smith a écrit :
> Stephen Adler wrote:
> > I disagree, may new features which make it into the kernel
> > are not 100% ready for prime.
>
> As a sysadmin, I have seen this with 2.4 in the stock, RedHat and
> XFS kernels. There are many things still incomplete and/or buggy
> with 2.4. I don't know how many times I've seen bugs reported on
> this list that end up being general kernel 2.4 issues. This is to
> be "expected" with 2.4's relative "newness" and as more and more
> "diversity" in systems using 2.4 find these issues.
>
In fact 2.4.0 does not mean that suddenly kernel has become flawless.
It means that the kernel enters
in period of general test. The real point where an even version is
ready for prime time is not 2.x.0 but the start
of development for new version. This is when Linus and others feel
confident enough about the kernel for not
needing the whole team watching for bugs. And ideally kernel should
noyt be included in distributions before this point.
>
> XFS is already proven on the Irix platform. And the structure is
> not changing (unlike ReiserFS, although I heard Linus made Reiser
> stick with a fixed structure before 2.4 inclusion?). IMHO, XFS is
> as ready for "production use" as ReiserFS -- more so if you consider
> the fact that the kernel must be patched so ReiserFS can work with
> kNFSd.
>
> [ Side note: I would love to see Ext3 put into the stock 2.2 tree
> as it, in v1/full-data-journaling mode, has been more stable than
> ReiserFS for me. I have been running with Ext3 on high-activity,
> production CVS/SMB/NFS _servers_ for almost a year now. ]
>
> Now being a non-developer, anyone should _rebuke_me_ if I am
> incorrect anywhere here. I'm just speaking from end-user/sysadmin
> use.
>
> > As a matter of fact, the only way you can make something
> > 100% ready from prime is to get everyone to use it on the
> > internet and then you will find all the bugs.
>
> Again, IMHO as a non-developer, XFS is at the point where it should
> be included in the stock kernel -- at least as much as ReiserFS
> should be. This *MAY* be a "flawed" viewpoint, and both should NOT
> be in the kernel and I'm surprised ReiserFS went in on 2.4.1 -- but
> I'm *NOT* going to "second guess" others who obviously know more
> than me about the code anymore than I have ;-PPP. But I see that
> XFS patches against the stock kernel well, the XFS developers are
> maintaining an up-to-date kernel tree in CVS themselves (because it
> patches fairly easily), and offers an existing implementation that
> works on other platforms (unlike ReiserFS).
>
> > I think reiserfs is a very good example. its in the official kernel
> > tree even though there are known corruptability issues.
>
> I don't know if I would go that far, but it does have some kNFSd
> issues (if not just interoperability, but performance).
>
> > Redhat then puts it into rawhide to let us bleeding edgers kick its
> > tires by building large file systems with it.
>
> Well, if it's in the stock kernel, RedHat's not going to yank it
> out. Frankly, I'd like to see Tweedie's knowledge put to use on
> XFS, but I'm not his boss, nor do I know what he's up to. ;-PPP
> Ext3 on 2.2 is great, I use it liberally, and I'm sure Tweedie would
> love to get Ext3 out on 2.4 ASAP. But maybe it's time we look at
> XFS as "our future" for Linux? Disclaimer: Just my $0.02 on that
> -- feel free to *SMACK* this "non-developer" silly. @-P
>
I think it would be great evil for Linux if instead of picking best
filesystem to replace
ext2 we selected one because it is most compatible. Migrating is a one
time affair (and there
are plenty of occasions it will have a zero cost like when you change
your disk) but using an
inferior FS means inferior performance for life.
>
> > Once a piece of software reaches a pleateau of stability within
> > the developers circles, you have to let it out for the rest of
> > us to use. As from what I read on the mailing lists,
> > xfs has reached a stability pleateau on par with reiserfs and thus
> > should be let into the kernel tree.
>
> That was my point. Although if ReiserFS isn't "really stable
> enough" to be in the kernel tree, what good does it do to put XFS in
> as well (two "wrongs" don't make a "right," eh)? I don't know the
> circumstances, but I felt like ReiserFS' inclusion in 2.4.1 was more
> political/demand than technical. But I'm not going to comment on
> that anymore since I'm probably making it in great ignorance.
>
> > OK, off my soap box.
>
> Yeah, me too.
>
|