"Nathan Whitehorn" <nathan.whitehorn@xxxxxxxxxxx> wrote: >Orange boxes notwithstanding, I still believe integration into the >kernel proper is a mistake, because it (1) requires our flexibility in >the long term, (2) splits up our code, and (3) makes networking a >special kind of kernel module, which , in my opinion, is a mistake in >terms of kernel design. While it may be a cleaner design for a kernel to not have any networking functionality, putting it into the kernel might be a nice idea anyway. Making networking a special kind of a kernel modul is perhaps the biggest hack, we should not do that. Either we keep the current way, doing everything using open, close and ioctl (which does not look very much like a hack when your kernel has a working select, which the new kernel has), or we add all functions to the kernel, not only socket, but also everything else, like recv, accept, bind, .... The first approach has the advantage that that you can remove networking from the kernel without needing a new kernel, the second has the advantage that your kernel has networking functions (but perhaps not all, what about the name resolution and other things, do we really want *all* network functions to be in the kernel. Needs to be done for full, useful support). I'm not sure. Has nobody else here on this list an opinion about this? Marcus