RE: [TCLUG:17072] Rebuttal

["Eric Hillman" <ehillman@cccu.com>]

> I don't know a whole lot here, actually.
>
> But if I want to change workgroups from SIGFS to TCLUG,
> or I want to leave the SIGFS workgroup, and join a domain
> (.sigfs.org), I have to authenticate to a PDC (I think?), and so
> it does weird and funky things.  Again, I'm not an NT expert,
> especially on server issues.  (this is like changing the hostname -
> which I think you have to reboot to do in Linux.  Maybe not?)

Nope.  There's *nothing* network-related in Linux that requires a reboot (except
for actually yanking out the network card, I suppose).  Networking is a set of
services that can be stopped & started independently of the Linux kernel.

I have never, ever had to reboot a Unix machine except when changing hardware,
or installing a new kernel.  It just doesn't happen, because of the way that
Unix is designed.  The notion of "rebooting for changes to take effect" is pure
DOS legacy.

> And yes, you don't need to reboot to change IP addresses.
> In fact, many things in Windows said they needed a reboot,
> but really didn't.

Rebooting to change IP addresses was addressed by one of the NT 4 service packs,
but the original 4.0 release required a reboot to change IP settings (including
DNS!), screen resolution, and all kindsa stuff.  Windows has gotten a lot better
at this in the past 5 years (largely, I think, due to pressure from Linux and
other Unix platorms) but no service pack could change the fact that there's way
too much networking, video and general stuff in the kernel.

I don't fully understand why you have to reboot in order to change your computer
name or workgroup (and, at least on my NT4sp6 box, you still do), but I suspect
that it's once again that poor seperation of the networking from the kernel.
Whereas Samba (by necessity) seperates SMB/NMB information from other networking
& OS information, with Windows it's all part of the same big ugly snarl of data.
MS actually advertised this as an *advantage* -- "tight integration of the
network with the kernel" -- which is, in my opinion, a load of smoke.  Putting
your networking code in the kernel doesn't necessarily make it run significantly
faster than just having a tight, fast kernel that allows external processes to
do their I/O in a timely fashion.  Instead it gives you a huge, bloated kernel
(because *everything* turns out to be "tightly integrated" in NT4 -- the GUI,
system logging, etc.) and turns patches & upgrades into a freaking nightmare
from a programmer's perspective.  Most importantly, it means that no matter what
your server's actual job is, it's *also* being forced to deal with the GUI, the
network, and all that other crap that got packed into the kernel every minute
it's in operation.  And a bug or misconfiguration in any one of those subsystems
can bring the whole machine to its knees.

Maybe this is different in W2K -- I haven't paid much attention to the new OS
yet.  I suspect that if it's anywhere near as stable as they claim, they must
have been forced to offload some of this stuff off of the kernel and let it run
in its own space.