The NSF is looking for a newer, better, shinier Internets.
In terms of underlying protocols... IPv4 is pretty darn good, in its place, keeping in mind that it was designed for a more civilized age, and that its architects weren't really expecting to see billions of nodes on the global network.
Then we have IPv6, a bloated monstrosity which allows networking every electron in the known Universe, and then some. While it has a great superfluity of addressing capability for the Internet of Doohickeys, it's fantastically complex, and enough trouble to implement and use that, after all these years, it's still largely failing to catch on.
So... maybe fall back to something in between? How about a cleanly layered protocol (which TCP/IP never was), with enough addressing for all the plausible needs of a medium-sized planet (64 bits ought to do it), and the IP- and TCP-analog layers being about as simple to implement and use as v4? Include hooks for security layers, QoS, and so on, but don't require those features as part of the baseline.
And: be mindful of overhead, especially since much of the expanded addressing is needed to deal with tiny devices with limited bandwidth. Look at how poor 6lowPAN has to break IPv6 packets into many little 802.15.4 packets, just because the header is so huge. Including a sub-IP protocol layer for channels with tiny packet sizes might be nice, too.
But, with gummint grants being involved... expect bloat.
Comments