Hi board,
I'm investigating how to configure VPLS to create a redundant connection to a CE in my testlab. The CE is a simple HP managed switch not running stp.
Site 3 is single homed to PE1 and Site 6 is multihomed to PE2 and PE3.
PE1 and PE2 are MX80's and PE3 is a MX960 all running junos 14.2
First question I have is that a lot of docs say that you should implement stp on the MX's that are configured as being multihomed. I didn't do that, so I was expecting to see a loop. However, it seems VPLS has a backup/primary role and the PE that was primary, was forwarding traffic. My tests showed that the backup PE was not forwarding any traffic.
So if this is the way VPLS works, why do you need stp???
Next I run into my issue: I will have active components in between the CE and the PE's. If I pull the link to the active PE, the link on the PE will stay up and the PE will stay primary. The backup will not take over and we have an outage.
How do you deal with this???
The only situation that I've gotten to work, is to configure PE2 and PE3 in different sites and remove the multihoming statement. Set the CE-facing interfaces to encapsulation ethernet-vpls and enable stp on the CE (HP switch). This way the HP switch will see his own bpdu's and determine what uplink to block.
Another thing I read, is that the PE routers participating in the multihomed site, should have a unique RDto help with faster convergence. I don't really get this, because I thought RD's only mean something to the local PE router to make sure entries are unique in the BGP table. I don't see the relation with other PE's.
Please help!!