[CentOS-devel] Too many kernels?

[Bogdan Costescu]

On Thu, 3 Jul 2008, Axel Thimm wrote:

> I start to think whether these kernels are indeed being all used to
> the extend of justifying full kmdl support. Maybe it would make sense
> to keep the full last series (2.6.18-92* above) and the highest one
> from the series before (2.6.18-53.1.21.el5 and 2.6.18-8.1.15.el5).

While Red Hat has promised initially a stable ABI within a version 
(including all updates), this was not always the case - or maybe it 
isn't clear to me what this promise means. However, the breakages are 
much more likely to occur when major updates (like RHEL 5.2) come 
while the minor kernel updates (like 2.6.18-53.1.19.el5) in between 
usually contain only security fixes (please note my usage of "usually" 
:-)). So I find it very likely that a module version compiled for one 
-8.1.x kernel will work on all other -8.1.x kernels and less likely to 
work on a 53.1.x one. Based on this assumption, keeping one kernel 
module for each major update and using weak-updates for the minor 
updates would probably work in a large percentage of cases.

For CentOS in particular, I can see that older kernels are actually 
not available anymore on mirrors, neither as separate packages nor as 
part of ISO images. The only way that I can think of to get those 
older kernels is to install from ISO images written some time ago on 
physical CD/DVDs - but these ISO images are only released for major 
updates, so only one kernel version per major update is actually 
available for those who choose this installation method and don't want 
to update to the latest and greatest. For this case, again, having one 
kernel module for each major update seems to be enough.

Disclaimer: I don't have any connection to Red Hat and I'm not a 
CentOS developer.

-- 
Bogdan Costescu

IWR, University of Heidelberg, INF 368, D-69120 Heidelberg, Germany
Phone: +49 6221 54 8869/8240, Fax: +49 6221 54 8868/8850
E-mail: bogdan.costescu at iwr.uni-heidelberg.de