[OSPF/LDP/PW Fast convergence] ASR9k/ME3800/ME3600

Question

Hello, Netpro:
I'm having a bit of a problem with OSPF/LDP Pseudowire fast convergence.
I have a test lab scenario with 2 ASR9000, 1 ME3800X and 1 ME3600X.
They are connected in an L3 MPLS ring, like this:
ME3800X---[a]---ASR1--[b]--ASR2---[c]---ME3600X
  |                                                                      |
  |--------------------------------[d]-----------------------|   
The MEs are connected back to back closing the loop.
The ring is configured for: OSPF fast convergence timers, LDP session protection, MPLS LDP  sync and BFP for OSPF on all interfaces.
For testing purposes I've configured a pseudowire between the MEs and forced the path via OSPF to go through the ASRs.
My problem is that I do not get the same responses when breaking the ring in different places. For example: if I break the ring in (b) or (d) i get ~100ms loss (when breaking the d connection, I change OSPF so that the PW takes the direct route between MEs). If I break the ring in (a) or ( c ) I get ~ 500ms loss.
Also for testing purposes, I ran a similar test but with regular L3 interfaces ( NO pseudowire) between MEs and for all scenarios i get 60-100ms. So, basically I'm assuming OSPF is converging like it's supposed to and the issue is within LDP.
Any ideas?
Regards,
c.

Answer 1

Yes, I did test traffic engineering, but there's a bug (on 3600/3800) that when a PW changes paths, it will stop passing traffic, so that part is stuck.
Is there an OSPF problem with ASR? I didn't understand very well your comment
My description of my scenario is just for simplicity purposes. The real question here is, what is going on between an IOS box and an XR box that when you cause a fault in between those 2, LDP recovers in 500ms and when you cause a fault in same-OS boxes, recovery is 100ms.