For the Cisco folks

As someone who has watched communications from a child with 90 watts of morse code to now, watching the evolving intelligent network, here is a rule.

The end points of data have become more intelligent at buffering and more complex at display, even as the long distance links grow in speed. Moore's Law balances the link and terminal.   Hence,there was always a barely noticeable pause that was available to the network, not a statistical gain, but the ability to manage end to end, or partially, out of band. When I was there, it was the ability to interleave advertisement into video, locally, along the net.  Moore equals out and keeps the network gain.

Today, it means that once the interleaved packets are separated, they can be routed onto slower, serial links, just note their arrival order. In that case you have the distributed block problem, but the packets are not that disordered.  Different risk processors independently routing them back onto the local links, in serial fashion.  But you have the latency to do simple block reorganization.

The distributed router can intermingle with the distributed database, which may have authorization to take some stream of IP for local consumption.  This can be done without the URL, local re-routing, as long as network contiguity protocol is met.  We get a local namespace, and Cisco leads the way. In some cases, congestion just pumps into a FIFO disk, for the moment; simple re-directs that are easily managed in a higher level language. Then, make your router class accessible in, say, python.

Then, over the shared, local serial link, one application imports the distributed routed and

for ip in the_stream(get_google_search,list_of()

Now this iterator will be issuing routing slips on set up, __init__ the iterator. All parties have one cisco driver in the python interpreter, a scremer, it can grab packets by IP, and.....do cisco stuff...