BNG geo-redundancy IPoE via VPWS observation and notes

Benjamin Bartsch · ‎03-03-2016

We can't use MC-LAG due to access equipment limitations.

We don't want to use MSTAG because...

Redundancy is not a switch you flip on or off - it takes a lot of work and coordination.

Redundancy introduces additional 'what ifs' - you create many more additional problems to fix (split-brain, routing issues, convergence times, etc) than you solve (node failure).

Routing is tricky if you want 'seamless' redundancy - bone up on your CCNA routing path selection criteria.

You have to coordinate several event to happen at the same time - pseudowire failover, routing failover, BNG master failover. Tracking objects are your friends.

Failover and failback do not always work as expected.

A huge thanks to Mr. Xander and Mr. Aleksandar for helping me hack through the geo-redundant BNG configuration. It's not perfect - it's like the Big Mac you see on TV versus the Big Mac you get in real life. But it's still works pretty well (who doesn't like a Big Mac?).

Benjamin Bartsch · ‎03-04-2016

Failover works fine in one direction, but not the other. When the master BNG node is down, the slave gets promoted to master (takes about 30-35 seconds). The subscriber sessions move over just fine at that point.

However, when the original master comes back online, he goes to slave, which blackholes our subscriber traffic as they go to nowhere.

We do a manual switchover from the wrong master node, and it switches over just fine (the expected master and slave routers assume their correct roles). However, it takes about 7-8 minutes for the subscriber sessions to swing over. We disabled all hold-timers.

TAC SR 638276513

xthuijs · ‎03-05-2016

Hi Ben! thanks for the earlier comment, haha, good think is that geo red doesn't make you fat at least :), it probably makes you lose weight from all the worries :)

Let me have a look at the tac case also and involve some BNG geored dev to have a look along and recommend.

Just for info, I wanted to pass on some known limitations we know of.

One of the things that comes to mind is the MTU/MSS between the BNG pairs, we are limited there, like BGP route updates how many updates we can pack in a sync over ICCP.

cheers!!

xander