In reply to Michael Gettes comments on the Backbone.
 
The problem in our neck of the woods is that YALEVM is not on the
backbone and is a major hub node.  This means that distribute jobs
have to go down one of the spokes and then back up to go down the
others.  However, I don't think that the node that has the most
nodes behind it would be the best choice either.  Presumably, the
traffic over a link is related (to some extent) to the number
of nodes behind it.  So since the jobs have to go down and back up
one of these links, perhaps the one with the fewest nodes behind
it (and presumably the least traffic) would be the best choice.
 
Actually, I do agree that our node is a poor choice in this instance
since the traffic on our link is very heavy due to a server here
that creates a lot of traffic.  Another reason is that due to some
problems Yale has been having with their line controller, they have
been draining our link at night.  In any case removing any node from
the backbone can only increase traffic.