Hi Misc,
I'm thinking about the best way to accomplish a redundant BGP setup with two
boxes running OpenBGPD.
I know there are 2 ways that are frequently suggested:
First one: active / passive setup with depend on. Bad because sessions are
reset, and bad as well because one router idles while the other one has to
deal with all packets
Second, better one: one carrier per router and linking them together with
OSPF and iBGP as discussed in Claudio's papers
But there are still a few things I don't like about the second way:
I would prefer to connect both carriers to both routers so that still both
carriers remain active when one router fails.
In addition to that I need either 2 extra machines for running OSPF or
traffic might flow twice over one single router which decreases pps
performance.
Now my idea looks like that:
carrier1 carrier2 carrier1
carrier2
| |
| |
|em0 |em1 |em0
|em1
|metric50 |metric100 |metric100
|metric50
| |
| |
----------------------
-----------------------
| |
router1
router2
| |
|em2
|em2
|carp with arp
|carp with arp
|loadbalancing
|loadbalancing
| |
coreswitch1---------------------------------------------coreswitch2
| |
|RSTP
|RSTP
| |
------------------------switchN---------------------------
|
serverN
My targets are:
Quite evenly load balancing to reduce pps per router.
Full redundancy - carrier, router or coreswitch can fail without service
interruption.
Avoid additional routers running OSPF.
So finally here come my questions:
Is this idea that I'm figuring out possible or have I missed something?
Has anyone set up such an architecture before?
Any hints or suggestions for improvements?
Do I reach all my targets: can I avoid OSPF and master/slave operation that
way while still having full redundancy without resetting sessions?
