[openbeos] Re: Networking : R5 or BONE
- From: François Revol <revol@xxxxxxx>
- To: openbeos@xxxxxxxxxxxxx
- Date: Sun, 27 Jan 2002 06:59:28 +0100 (MET)
En réponse à John Gabriele <jandl@xxxxxxxxxxx>:
> I'm maintaining the faq and have been watching this space to update
> that particular faq item. The reasons I haven't yet changed it are:
>
> 1. Which is it? Zoidberg's user-land stack? Travis's kernel-space
> stack?
Maybe both, since Travis me said he was coding one, but didn't want to lock us
on his implementation. So maybe we can get both, just like BONE ppl can switch
back to net_server (does it work btw ? I never tried it =) )
> 2. I don't understand what a "network stack" is.
It's the software layer that manages the network cards (including ppp over
modem), deals with networking protocols like IP, TCP, UDP... (IPX ? :P),
and delivers the data either through the posix socket interface, or the NetKit
OO interface.
> I further don't understand exactly what having a kernel-space network
> stack entails.
It means all the thing runs as kernel level drivers and threads, that is
in the kernel space (except the Net Kit, which anyway is a userland C++
library). That brings lots of good things, as well as some downsides as someone
else already described here (+: no context switch, less data copy
-: stability ? ...)
> Does it mean that there's gonna be the regular posix sockets calls in
> the kernel kit (and that, when called, they will execute in
> kernel-mode)?
Yes. But, anyway even making a net_server implementation wouldn't prevent us
from having sockets as files (that is file descriptors, which are handled by
the kernel), just we could use a kernel socket driver or so which would call
net_server. And that is what we agreed I think on a mandatory point.
>
> If OBOS uses what Travis puts into NewOS (this network stack), will
> there still be a net_server?
Why not ? ;)
François.
Other related posts:
