Bradley D. Thornton wrote:

> Okay, certainly not a thousand; several hundred; and at least 30,
> maybe 40; and in that order.
> I know in practice people build kernels everyday and tote them around
> with them.
>

Which is why we need a Survey/JTA to see how much.  I have a feeling it's
at least 0.3%, maybe closer to 5%.


> Perhaps you're only considering the more general purpose distros with
> support contracts?


You mean Sustained (Enterprise) Linux distributions?  The ones sustained
for 5-7 or even 10-13 years for a fixed Application Binary Interface (ABI).

Other than prepping the kernel for adding modules, it's kinda inversely
proportional to the reason you'd want to run one.  If one is building
kernels, and even core software, it kinda defeats the purpose of having a
fixed ABI Enterprise distribution.

And for embedded or IoT devices that are consistently updated without any
need for a sustained ABI, most people aren't running a Sustained Linux
distribution.

The kind that led me to completely eradicate RHCE
> from our enterprise at Medata when they told us that a kernel compile
> that we needed for our software would void our support contract?
>

Well, Red Hat (RHEL), SuSE (SLED/SLES) and even Canonical (Ubuntu LTS) have
various guarantees on ABI compatibility.  So other than prepping the kernel
tree for building additional modules/subsystems to load, changing the
core/stock modules will result in such.

Perhaps you're not considering Arch, Gentoo, Slackware, and many other
> distros used on both the desktop and in the enterprise.
>

That's why I don't like the term 'Enterprise' but more 'Long-Term
Sustained.'  It costs Red Hat and SuSE and inordinate amount of sustaining
engineering to backport and keep the long-term ABI, as even Canonical is
founding out as its Advantage program moves from 5 to 7-8 years.

But if one's enterprise is rebasing within a year, or even continually,
it's really at odds with the model.  The classic Ian Murdock discussion on
lifecycle applies, which explains Debian as well.  The BSDs have used a
'ports' approach over the years too, although it's ABI/API doesn't change
as much as GNU/Linux.

A mention? No, I think this is a core competency and it would take two
> hands just to count the people I know who do kernel compiles each week
> - - obviously, my circle of geeks seems different than yours, but the
> suggestion that, "hardly anyone builds a custom kernel" is hardly,
> even remotely accurate.
>

Which is what a Survey/JTA should identify, at least if it surveys a good
portion of the hosting and custom software development market.

It's getting interesting nowdays because Red Hat has more of a 'core' ABI
for hardware/base software, but then an 'Application Streams' (aka Fedora
'Modular') approach to non-core libraries and software.  The idea here is
that the ABI is for hardware and base software that has to be fixed, but
everything above it should 'Modular' (Fedora term) or 'Streams' (Red Hat
term).

Containers then take this to a new level, and Red Hat actually had
OpenShift before Kubernetes, Docker and even LXC existed.  It came out of
the need for JBoss to run different, multiple libraries and language
support, especially between Dev, Test, QA and Prod releases, and "Software
Collections" (SCL) wasn't 'cutting it.'

But the kernel and core ABI is still fixed, even with RHEL 8 'CoreOS' now
at the heart of OpenShift Container Platform (OCP) v4.

If someone comes to me interviewing for an admin job and they only
> have a *concept* of a kernel compile or toolchain, I can assure you
> that they only have a concept of a job.
>

Depends on what the client/customer needs.

Configuring and compiling a kernel has never not been a core
> competency, and is also one of the things that has always set LPI
> apart from silos that offer their own certs to suit their narrative of
> what an engineer should have in their magic bag of Trix.
>

All the more reason to ensure it's in LPI, especially if we can justify it
via a Survey/JTA.

Regardless, I'd argue 'prepping' the kernel is still needed just to build
modules and, even more so, entire, modular subsystems that may not be
included in the 'stock' distribution.

Thanks for asking, I'm really glad you did too :)
>

That's what this list is for.  Now if we can quantify it with a Survey/JTA,
even better.  The good news is that it's already in the objectives ...

So it's more about justifying its removal than its continued inclusion ...
at least that's what I think (don't quote me).

- bjs
_______________________________________________
lpi-examdev mailing list
[email protected]
https://list.lpi.org/mailman/listinfo/lpi-examdev

Reply via email to