
|
[openbeosnetteam]
||
[Date Prev]
[08-2006 Date Index]
[Date Next]
||
[Thread Prev]
[08-2006 Thread Index]
[Thread Next]
[openbeosnetteam] Re: r18408 moved to network
- From: Ingo Weinhold <bonefish@xxxxxxxxxxxxxxx>
- To: openbeosnetteam@xxxxxxxxxxxxx
- Date: Sun, 13 Aug 2006 11:56:57 +0200
On 2006-08-12 at 23:21:47 [+0200], Oliver Tappe <openbeos@xxxxxxxxxxxxxxx>
wrote:
> On 2006-08-12 at 20:39:32 [+0200], Axel Dörfler <axeld@xxxxxxxxxxxxxxxx>
> wrote:
> [ ... ]
> >
> > I thought net_server should only be needed if you need:
> > * inetd functionality
> > * dynamic configuration loading (node monitoring of config files)
> >
> > it could also contain any tray icon stuff.
> >
> > DHCP needs a running server, too, to see if a lease has to be renewed
> > whatever. For future zeroconf/Bonjour functionality, this probably also
> > need a server of some kind. Since I thought the stack should be as self
> > -containing as possible, having DHCP support built-in sounds not too
> > bad to me (especially with Bonjour like stuff to come).
> > But if you think that's insane, and net_server should handle this, I'd
> > be okay with this, too.
>
> Well, I was just curious as to why you thought DHCP was required to live in
> the kernel >:o)
>
> Having DHCP live in userland would make it possible for the user to replace
> it with something else, though. And I personally would prefer it if the
> kernel wouldn't create any (UDP-) sockets by itself, but if there would be a
> clear border between the kernel providiing the basic protocols and the
> userland creating the services and sockets.
> But that is just my personal taste, of course.
Please also keep in mind that there applications like networking file systems
which preferrably live in the kernel, too. I hope you're not moving so much
functionality into userland, that those become a pain to write. Not having
sockets sounds a bit like that would be the case.
CU, Ingo
|

|