You need to make sure that each PE that is advertising the prefix has a unique RD.
You then also have to make sure that the RRs are enabled for multipath. This will mean that the RRs will pass both routes to PE3 so that it can either select the single best path - will come down to IGP cost - or enable multipath so that PE3 will use both egress points.
That should do it. If you enable multipath on PE3 but not on the RRs then PE3 will never get the chance to see both paths.
... View more
"ip source-route" does NOT mean that you are routing by the source address. The clue is in the name "source-route" as per earlier comments (13 YEARS AGO) means that the routing is determined by the source DEVICE i.e. the sender of the packet. It has nothing to do with the IP address of the sender. The sender determines what route the packet should take (can be based on anything - automatically discovered or prescribed) and then compiles the list of intermediate routers that the packet must go through to get to the destination. The list is then inserted into the IP packet (in the header) and since each router sees that the route has already been determined by the sender it then honours (as best it can) the routing by forwarding the packet to the next router on the list. It is dangerous because hackers can use it to circumvent security policies if they know that diverse routes exist. Source-routing deviates from the destination-based routing paradigm. Token-Ring networks from years back made heavy use of source-routing (at the MAC layer)
... View more