>> Well, in an exciting week the code now runs in kernel space, with all >> modules being loaded at run time. The drivers are still the original >> be ones >> being used via the same interface as the userland stack, which does >> mean >> that our code can be built to run as either a userland application or >> as the >> kernel drivers! > >Excuse my ignorance on the matter, but a few queries.... >How does integration into kernelspace affect the way users have control >over the networking modules? It shouldn't have all that much effect. >What I mean to say is, if there is a problem, is it still possible to >simply 'restart' the net_server? Per Be, the only reason that there was ever an option to do so was that the net server sucked. :-/ >And how about stability... is it wise to move things >into kernel space when they're still untested... even when they are tested? Fair question. It is a weighting issue. On one hand, it is much easier to write the network stack. And faster. On the other hand, if the net stack crashes, the whole machine could die. >As much as we might like to think otherwise, there are plenty of >situations when things break, and killing the kernel team is probably not a >NiceThing[tm] Very true. And this is the arguement that people make all the time for microkernels. No matter what decision we make, someone will not like it. :-)