[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: does streamlining kernel config really not help performance?
- To: misc_(_at_)_openbsd_(_dot_)_org
- Subject: Re: does streamlining kernel config really not help performance?
- From: Ben Goren <ben_(_at_)_trumpetpower_(_dot_)_com>
- Date: Mon, 22 Apr 2002 14:41:13 -0700
On Mon, Apr 22, 2002 at 09:10:27PM +0000, Miod Vallat wrote:
> > This reminds me of something I've wondered about. This whole
> > config -e thing just feels awkward and clumsy compared to the
> > rest of OpenBSD configuration. Is there any fundamental reason
> > there can't be a file, such as /<kernel-name>.conf, that has
> > these options in them?
> And where on earth would you put such a file if you're trying to
> boot from the installation media and, for some reason, you
> encounter a problem that only boot -c can help you prevent in
> order to at least install the system?
I'm not suggesting this replace the current mechanism; only
that it augument it. There's no need to throw out something
which does work and which is, in some cases, very practical. I
_am_ suggesting that, for the majority of cases, it's awkward
and potentially confusing, unlike, say, /etc/rc.conf or
By all means, keep config -e. I'd just like to see something
additional that is a bit less complex.
For properly-functioning systems, if the file is in the same place
as the kernel itself, I can't think of any reason why the kernel
couldn't consult it early enough in the boot process--perhaps
right after the options controlled by config are read? As I say,
I'm no kernel hacker; if there are reasons why this couldn't work
or is a Bad Idea, that's fine.
With something like this a great many people who run a custom
kernel could instead run GENERIC. Those who are most prone to
compiling a kernel and screwing it up would have no remaining
excuse to do so. The problems they would create for themselves
(``I set BUFCACHEPERCENT to 90 because I have so much RAM and
now strange things happen!'') would tend to be immediately
self-evident and should be relatively easy to diagnose.
The ability to modify a kernel without having to recompile is
a wonderful thing. I just think that an additional level of
abstraction is called for.
[demime 0.98d removed an attachment of type application/pgp-signature]
Visit your host, monkey.org