[nylug-talk] Does anyone have a case why we should still compile custom kernels

[Michael K Dolan]

nylug-talk-bounces+mdoln=us.ibm.com at nylug.org wrote on 05/30/2007 12:01:53
PM:

>      Some colleagues of mine feel that custom kernels is a waste of time.

In 99% of business, production system cases, that is true. If you're using
Gentoo, rolling your own kernel is usually required but also rewarding -
doesn't seem to be the case here.

Typically you would only want to compile a customer RHEL kernel for IBM
hardware if you needed an extra 1% of performance squeezed out of the
system, for lightweight, purpose-optimized images deployed in a cluster
environment or if you had some specialty hardware or security requirement
that necessitated a custom kernel. Otherwise, rolling your own only
introduces long run headaches especially with kernel patches, updates, etc
over the long run. Unless you have a specific need for a custom kernel,
makes sense to me to go with the stock RHEL kernel.

> Note, we have a decent amount of Redhat systems running on IBM hardware.

So there should be no special hardware requirements for the IBM systems
requiring a custom kernel unless you have very unique cards that are not
supported in the generic RHEL kernels...

> As someone already stated, most of the drivers are built for certain
hardware.
> I personally feel that custom kernels are necessary.

Why?

> Well actually, I
> worked with a guy that compiled custom kernels for all the boxes he
built.
> I feel this is the best way to go.

Custom kernels may show off your expertise in building Linux kernels, but
the problems come into play when someone with less time or less experience
has to take over management of that system... it's like taking over
someone's pet PHP project with tons of uncommented functions... add in
multiple systems with different kernels and it becomes a nightmare.

>      This question came for the fact that we have a client that's trying
to
> compile his/her own kernel(This person is having trouble with the
compile).

Why do they want to compile their own kernel? Just because?

> No one here wants to slave over fixing a clients mistakes.

Right. Nor should they unless you have a specific reason for this client to
build their own kernel. Your clients mistakes now will only get worse with
patches, drivers, etc over time... especially if they have many systems.
They could eventually end up with 10+ different variations of kernel
builds... it's tough to manage/support that environment.