BTW, is using just BGP really, really required?
What end result do you what to accomplish? Is the how it's accomplished as important as the end result?
Reason I ask these questions, I've seen proposed designs where some traffic has to go this way while other traffic needs to go that way, and when I ask why, the answer might be some traffic is more critical than other traffic (which is not unusual). But, if so, using QoS might be a better approach.
One simple issue I bring up, if some traffic is more important, and so you direct different kinds of traffic to different paths, what happens when one path fails? Not only do you lose that path's bandwidth, but now all traffic ends up on remaining path.
With QoS, you better guarantee more important traffic gets preferred handling, regardless of which path it takes.
A real-world example, to which I was personally involved, a large international company I was working for, wanted all remote sites to have dual WAN paths (and redundant hardware), as it was considered business critical that a site didn't drop totally off-line due to some single point of failure.
Great goal, but also great cost for dual international leased lines. So, to reduce WAN costs, leased lines were replaced by frame-relay, then by ATM and then by VPN MPLS. But all the private network WAN technologies were often much more expensive than an Internet connection.
So, for "less" important sites, VPN across Internet was used to replace one of the private WAN links.
But, with that, a new design goal was: "important" traffic should still use private WAN link, "less" important traffic should use Internet VPN.
After about a year or two, our international HQ figured out some way to route different traffic types across different paths.
In the meantime, in our regional HQ, we deployed identical QoS, across both private WAN and VPN Internet, and let traffic use either path. For all practical purposes, we got same results regardless of path (even running VoIP on both).
When international HQ finally came up with the kinds of traffic routing solution, and I showed them over a year of no problem actual operational results, including dealing with link failures, and asked, what happens when your "important" path fails, it was decided to use the QoS approach, everywhere. (Oh, and since Internet VPN worked so well, it was used more often as second WAN link.)
