On the left coast, we peer with PSI. We buy transit from MCI. PSI
recently (about a month ago) installed a broken localpref to prefer
their private MCI connection. It doesn't matter what we advertise
to MCI (double-hop, origin incomplete), PSI still uses their private
MCI peering - which of course is full during the day, so PSI
customers get packet loss.
Naturally, the problem does not exist on the right coast
I've called the NOC about 20 times now, trying to get it resolved.
I've been promised, multiple times, that it would be fixed, only
to get told today that "it's a policy issue". What?
Not that I havee actually seen this problem because we don't buy transit
from MCI, but...
You may want to try adding a few additional copies of your AS to your MCI
advertisement "prepend as 65536 65536 65536" where 65536 is actually your
AS. It largely depends on _how_ PSI is implementing their policy. It
could just be that all routes learned from their West Coast router (i.e.
a Border Router) are given a lower priority than private connect routers.
I guess a PSI person would be much better at answering with specifics. I
know I have seen customers get traffic flows to jump through the most
insane paths just be adding god awful AS lengths to their announcements.