Re: x86 release builds are slow

[Andrew Doran <ad%NetBSD.org@localhost>]

On Mon, May 05, 2008 at 06:33:35PM -0500, M Graff wrote:

> Andrew Doran wrote:
> | I think that's simply fallacious. Yes, our kernels are too fat and we are
> | working on that, but you would be hard pushed today to buy a new PC
> off the
> | shelf with less than 512MB of RAM. We're speaking of virtualization, not
> | embedded systems or constrained systems that can just about support
> one OS.
> 
> Fully-loaded XEN DOMU kernels also boot slower.

If you are rebooting your DOMUs all the time, presumably they are not in a
production environment, so you have the luxury of time. In that case why not
build your own tailored kernels, or even better, spend some time to optimize
the boot process so that it's quicker?

> You would be hard pressed to buy an off-the-shelf x86 system that cannot
> compile NetBSD DOMU in under 3 minutes.  We're not talking about user
> time here, we're talking about automated build time.  Who cares how long
> that takes, really?  I care about boot speed, and about binary size.

Sure, I got that. I understand that fewer+larger kernel images doesn't fit
with your meticulous set of requirements, that you don't see the efforts
being made to modularize the kernel, and that you don't see the possiblities
that a modest level of standardization brings.

> Why should I care how long it takes to build 11 kernels on beefy build
> hardware?

You probably shouldn't care about that, but I can't speak for you. ;-)

Cheers,
Andrew