On Sat, 2006-08-19 at 10:25, Timothy Murphy wrote:
> > Centos 3.x is still a good server choice. I've always considered
> > linux kernels to be stable around version x.x.20 or so.
>
> Which kernels have you had problems with?
> I've compiled all, or nearly all, the 2.6 kernels.
> IIRC, there was a problem with one, which was quickly corrected.
> Otherwise all have worked perfectly.
I've had some trouble somewhere with about every version
from 2.6.13 up, although I'm not sure if it has been the
kernel itself or fedora's initialization of devices. Not
recognizing firewire drives went on for a long time, then
there was a problem with IBM's on-board SCSI, then the
broadcomm gigabit ethernet.
> You keep making these statements about "instability",
> but what exactly do you mean by this word?
It has two different but related meanings. One is pure
rate-of-change. The other is detrimental changes where
something that previously worked is broken in an update.
The latter is the meaning that people usually care about,
but if you have some experience you will realize that every
change carries a risk of error and unforeseen complications
and only the first meaning can be observed ahead of time.
If the changes could have wide testing before release, you
could somewhat separate the two concepts, but in fedora they
don't, and if you follow this list you will see many instances
where an update breaks things. Currently you'll see the most
of these regarding Selinux and SATA controllers/drives because
there is a lot of change happening in those areas but there are
many more subtle things and changes in the kernel that don't
technically break things but require corresponding changes in
the rest of the distribution that are not necessarily coordinated.
An example would be the udev scheme which is now wildly different
from the old way of handling devices. One of my machines uses
amanda to back up several other machines to tape and has always
used /dev/tape as a symlink to the /dev/nst0 device as the target
in the configuration, but this no longer exists in FC5. The
'mt' command still has a default of /dev/tape though and
commands will fail if you don't give it an explicit target
now.
> Windows and Linux used to be relatively unstable,
> but both have been completely stable, in the normal meaning of this term,
> for years, in my experience.
Either you have been lucky or you don't expect much then. I've
had both windows and linux updates break previously working
things in the last few years.
> When you use the term "unstable" would you mind explaining
> what you mean by it, please.
With a unix-like system, stability means that you can write some
scripts to perform certain functions and go away for a few
years doing nothing but system bugfix and security updates
and come back to find it still doing its job. That's worked
for me so far with RH 7.3 and CentOS 3.x, and not much else.
The need for the bugfix/security updates is the killer here
because other distributions have allowed additional changes
that affect behavior to be slipstreamed into the updates that
you must apply. In fedora these changes are an expected feature,
not a bug...
--
Les Mikesell
lesmikesell@xxxxxxxxx
