On 8/20/07, Andrew Lentvorski <[EMAIL PROTECTED]> wrote: > Bob La Quey wrote: > > NixOS and NetKernel are all I see out there (other than > > virtualization) that appear to me to be significant steps > > forward. > > NixOS seems like the right thing. I'm not convinced they're going about > it the right way, but they will probably lead they way for someone else > to get it right. > > NetKernel, I think, is probably not a significant advance in distributed > whatever. > > I have a real concrete test for whether something "distributed > programming advance X" works. > > Does it punch holes through NATs?
That is not what NetKernel is about. It is about moving REST down to the level of the chip with many CPUs. NetKernel is what we do about parallel processing when we have many cpus on a chip i.e. we turn it into load balancing. I think you miss the main point. > That simple concrete problem tells me everything I need to know. > > Hole punching either requires *really* good abstractions, or it requires > the ability for the low and high level portions of your framework to > interact cleanly. No doubt we need better abstractions. I do agree that the high and low level portions of the framework need to work together. > It also requires the ability to think about differences in latency and > bandwidth as well as how to deal with node failure. Right. This is _not_ about using tcp/ip on top of the chip. So it has nothing to do with NAT. NetKernel is about understanding that a URI is meaningless without a context for resolving it into an address but DNS is only one of many contexts. A context and transport that takes into account the issues you bring up is needed to map REST style ideas onto the multi-cpu chip. So NetKernel adds transports other than tcp/ip to deal with onchip (and near chip) communications. > Everybody who does "real" distributed stuff (Skype, Microsoft, Sony, > Nintendo, Blizzard) always seem to manage to punch holes through NAT's. > The poseurs (BitTorrent, aMule/eMule/eDonkey, JXTA, etc.) never seem > to be able to pull that off. Again the innovative parts of NetKernel have nothing to do with TCP/IP or NAT. They are operating well below that level of the system. BobLQ -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
