🚀 go-pugleaf

If two nodes are K hops apart, there are K bits on in their routing mask
(the bit-mask generated by the from ^ to computation).

By always looking at the L.S.B of the routing mask, you're choosing a
single fixed path for communication between each pair of nodes.

If you choose one of the on bits at random, or if you consider every 'on bit'
and send the message out the channel with the smallest message queue, you'll
have a better link utilization. From simulation data for a network switch 
which allows all links to be utilized all the time, this packet routing method
will buy you some improvement (about 10%) when queues form. 

However, the Intel & Ametek cubes allow a single (or in the case of Intel,
maybe two) concurrent transmissions.
	In this case, it doesn't matter which routing method you use. Link
utilization is not the question: it's memory traffic within a node.

I'm not certain how many channels Ncube will fire up at once.