A good Linux kernel vintage?

[Oleg Goldshmidt]

On Wed, Nov 16, 2011 at 5:19 PM, Eli Billauer <eli at billauer.co.il> wrote:
> Hello,
>
>
> I'm running Fedora 12 on my main computer, with no intentions to upgrade the
> entire system (as I have a lot of non-distribution software which will be
> headache to reinstall). My choice of distribution in indeed questionable,
> but not the issue.
>
> I'd like to upgrade my kernel to anything > 2.6.36. But as many of you know,
> Linux kernels are a bit like wine: You know if you got a good one only after
> opening the bottle and waiting a little.
>
> So can anyone point at a kernel version (possibly flavor) which is known to
> be a successful one? I'm not looking for answers such as "I'm running kernel
> X.XX.XX with no problems". You may have problems you're not aware of. For
> example, I want to leave 2.6.35 because of that pretty famous system freeze
> under intensive disk load.

Well, at this level of fine-tuning maybe you should come up with a
list of "requirements" or "points of interest"? I mean, without
knowing what things are important to you people will be liable to say
"X.XX.XXX is great" or "don't touch Y.YY.YYY" while the pros and the
cons will be quite irrelevant to you. I run Fedora on my home (and
some work) machines, but none of them is stressed enough to struggle
with a particular kernel version as I update from time to time.
Therefore to me it sounds like you have more stringent requirement
that I do not understand.

In my case, the biggest issue recently was, I think, VMware
Workstation 7 not working with Fedora 15 kernels (I think 2.6.38+)
combined with unwillingness to pay - again - for Workstation 8. I
tried to patch the VMware modules by hand to make them work (I did it
on multiple occasions in the past) but then I decided that I would
need to do it for every new kernel version and it would be simpler to
just stick to the current Player that worked just fine. Obviously, I
don't know if it is relevant to you.

Baruch's suggestion of 2.6.32 has merits but does not satisfy your ">
2.6.36" criterion. While 2.6.32 is the stock of RHEL/CentOS (I have
"6" machines around, just checked), Fedora is ahead.

I cannot recommend a specific version, but my intuition tells me that
if you use Fedora on your main machine, with the ensuing expectations,
then maybe sticking with RedHat/Fedora kernels is a good choice. I
suppose it should be possible to update the kernel, even to one from a
newer Fedora, while leaving most of the userspace intact. YUM should
help you find out what dependencies will be dragged in. At the
major.minor.patchlevel granularity there are not very many choices:
Fedora 15 is at 2.6.40 now.

In addition, if faced with such question, I'd try to query
RedHat/Fedora Bugzilla for issues with each of the candidate kernel
versions.

Finally, a side note. I am reading your comment on "non-distribution
software that will be a headache to reinstall" as software that will
need to be recompiled/relinked with newer (shared) libraries, etc. The
software itself is probably under /usr/local or /opt, and hence
hopefully easy to tar/untar. I would consider creating a VM with a
newer Fedora (or whatever) and checking whether things still work
there. Maybe grab the missing/obsolete so's and put them in /usr/local
on a new machine. This may be too much work indeed, or it may be not
so bad. The point is there are ways to try non-destructively, in
principle.

-- 
Oleg Goldshmidt | pub at goldshmidt.org