[sclug] Recovering gracefully from lost interrupt

[lug at assursys.co.uk]

On Thu, 7 Aug 2003, Simon Huggins wrote:

> On Thu, Aug 07, 2003 at 12:17:41PM +0100, lug at assursys.co.uk wrote:
> > On Thu, 7 Aug 2003, Simon Huggins wrote:
> > > On Thu, Aug 07, 2003 at 12:54:59AM +0100, Will Dickson wrote:
> > > > Perhaps, although IIRC is was just 2.4.20; I can't remember whether it
> > > > was Debian's precompiled one or a customised one. (In the latter case,
> > > > I am familiar with the kernel compile process; I've been compiling
> > > > customised kernels successfully for several years prior to this
> > > > incident.)
> > > Distribution kernels are rarely the same as a particular vanilla version
> > > from kernel.org though.
> > Most distributions (Debian, Red Hat at least) include a significant number
> > of patches that fix stability and reliability issues with the vanilla Linus
> > (Marcelo) kernel.
> 
> Absolutely this is my point.  There could be any number of patches
> floating around in the Suse or Debian ones which cause the problem to
> appear or disappear.
> 
> > As these patches are picked by people more closely involved in the
> > kernel development process (and thus, probably more knowledgable) than
> > myself,
> 
> More closely than Linus or Marcelo? :)

In production enterprise environments, yes, probably. IIRC, Linus has even
put out "stable" kernels with fatal flaws before now...

> > I generally use distribution kernels on production machines.  I'll
> > only use custom kernels on non-mission critical machines, and even
> > then, I'll go to some lengths to include the same set of patches -
> > checking for obsolence if a patch clashes, and back/forward-porting if
> > necessary.
> 
> I agree that commercial distributions often do lots of testing and
> include fixes for bugs in vanilla kernels.  They do however normally get
> folded in upstream too eventually.

Sure - the bulk of the patches that RH include are usually moved into Alan's
-ac patches and thence into the stable kernel. But of course by then, other
bugs are known about and rough fixes and hacks for /those/ are circulating.

If you want the most stable/reliable kernel *today*, it's probably a distro
kernel.

If you want the most elegant code, sometimes at the cost of
stability and/or reliability, it's probably the vanilla kernel.

Y'pays yer money and yer takes yer choice!

Best Regards,
Alex.
-- 
Alex Butcher      Brainbench MVP for Internet Security: www.brainbench.com
Bristol, UK                      Need reliable and secure network systems?
PGP/GnuPG ID:0x271fd950                         <http://www.assursys.com/>