> > What about for those things that are hardly ever used in a production
> > server, like the floppy drive, but it would be annoying to have to
> > recomplile the kernel for :)
> >
> > > Production server which is fixed in configuration and stuffed into a
> > > closet, works best with monolithic kernel.
> >
> > But this isn't how many people use their linux machines.... is it?
> >
> > > Any lurking need to use kerneld is easily cured by getting enough
> > > memory.
>
> I've not messed with kerneld yet...but for my laptop with limited RAM, I
> can see a value to it. For my servers, which all have 32-64mb RAM, why
> bother with kerneld to save even a few hundred kb RAM. There's not a
> heck of alot on a server that could be moularized and left unloaded. The
> floppy code is about all that comes to mind. I suppose lp and serial
> could also...assuming they can be modules.
sigh... everyone's still missing the point (not all that uncommon on this
list). the point of modules, IMO, isn't really to save memory. even
with compiling almost everything but the bare essentials as modules,
modular code takes only 42 pages, or ~168k if i'm not mistaken. the
trouble of compiling things as modules (and yes, it is a LITTLE trouble
to do, just a bit tho) is not worth it. i mean, bash alone takes twice
that.
what it IS useful for, and, again IMO, underused for, is for distribution
of new kernel code. the mac filesystem is already distributed this way &
you don't have to rebuild the kernel just to add it on. dosemu also works
this way. it's really convenient (to me, at least) to be able to
install it and go, rather than wait for the kernel to build (which, of
course, takes forever now, even if you're only changing one option, but
that's beside the point). even if you can build the kernel in 10
minutes, you might not be able to take the machine down for even one
second to reboot (such as our work shell machine).
of course, this is all whimsy right now anyway. it's pretty difficult to
load a module compiled under patch level 143 when three releases later the
kernel interfaces will change to the point where the module is unusable.
versioning info was supposed to solve this problem i thought, but if you
make a function take an extra argument somewhere, that pretty much breaks
it. and we all know how the kernel developers LOVE to change the
interfaces without telling anyone... (ok, so it's documented in the
source & patches, forgive me if i don't scan each release meticulously
looking & trying to understand the changes)
shag
Judd Bourgeois | When we are planning for posterity,
shagboy@thecia.net | we ought to remember that virtue is
Finger for PGP key | not hereditary. Thomas Paine