[Novalug] email server locked up again (twice in two days)
kevin at pheared.net
Sun Feb 4 12:21:50 EST 2007
On Sun, Feb 04, 2007 at 11:52:00AM -0500, gregory pryzby wrote:
> > slow anything down. Loading modules only slows boot time, and not by
> it has to load it and that takes some time and extra memory. Not a big
> deal now that there are 100sMB and GB available, but when I had 16MB,
> it was a big deal ;)
The kernel has to be read from disk regardless of whether its parts are
in pieces or one large file. The question is whether you do that later
or all at once. I'm telling you that the difference is essentially
negligible. You would be optimizing the wrong part of the process.
It doesn't take much memory to load a module, but like you say, if
you're sweating an extra couple of MBs, you're probably running an old
system. In which case, you're probably too cheap to upgrade anyway.
> > much. The kernel doesn't have to load every module that has been
> > compiled, therefore occupying the same space in memory. So I don't see
> Corret, just what it needs when it needs it. so there is the
> realization that a module is needed, then find and load, and then reap
> the module when it isn't needed. Sure not a big deal, but it might
> make a difference over time. Why not just have what you need in the
Modules aren't unloaded unless you ask. They don't show up and
disappear on-the-fly. So there is no difference over time, it's simply
a matter of loading them at boot time.
Why not have the kernel support anything you might use? Also, as
modules, you can disable support for something by unloading the module.
Passing kernel parameters means rebooting. Not very flexible.
> It actually is larger memory footprint as a module. I would need to
> look up the exact reasons, but I do remember that was the case. Maybe
> that has changed in the 2.x days.
I'd be interested in knowing how much more memory is consumed by a
modularized piece of code. I'm going to make a SWAG that it's not worth
> Also, if I compile, I can make sure it is compiled for MY processor
> and takes advantages of the chip. This is huge for the x86 as kernel
> was compiled for x386. If I recompiled for 586 or whatever I had,
> there was a difference.
Any modern distribution will provide kernels built for 686, 386, etc.
For most applications you won't even notice the difference though.
> Sure I am talking about tiny bits of speed and memory, but for the
> super-tweaker, it makes a difference.
> That is ONE reason I just use stock kernels-- I don't care enough.
> --greg 'former super-tweaker'
I guess my point in this thread has been that 'super-tweaking' isn't
likely to make a system more stable, may not provide any appreciable
benefit, and is difficult if you don't want to get into the details of
how the operating system bootstraps itself, etc. Not to mention that
the average computing resources user doesn't necessarily have a CS
degree, and honestly shouldn't need one, but will need one to make many
of the choices provided in make config.
Those of us who do like to tweak, will tweak, but it's not going to be
helpful to suggest to everyone that has a problem with a system to build
their kernel from scratch. That will just add entropy.
More information about the Novalug