2

I am trying to come up with design for my network. Need your suggestions:

Two options:

N1: Two OpenBSD boxes, each connected to ISP, BGP configured with providers only. Both boxes connected to LAN, CARP configured between two boxes with virtual IP that will jump to another server if BGP link on active node goes down.

Servers in LAN (RedHat, Debian) have Virtual IP as default GW.

N2. Two OpenBSD boxes, each connected to ISP, BGP configured with providers and between routers. OSPF configured on both routers.

All servers in LAN have OSPF configured and route traffic according to it.

I personally like second option, since it promise faster fail over. and if everything works as supposed to be more straightforward setup. But I do not know how well OSPF works on Linuxes.

Let me know.

2 Answers 2

6

Honestly... my true opinion on the subject is that you let router appliances do the routing... and let servers be servers. I.e. do not put OSPF/BGP on "servers" ... let the edge-routers handle BGP to route properly. If you're trying to cluster two (or more) servers using the same IP... you should look at the many clustering strategies on a shared IP. Having OSPF on the servers themselves will not guarantee a faster "fail-over" ... in fact... it may be even slower.

3
  • No. only OSPF on servers. so they know which router is active now. But i see your point. Jul 7, 2011 at 21:55
  • Any routers (that I know of) that supports bgp... is capable of having either a shared IP between them... or automatically removing/assigning an IP & MAC address to an interface when the other side goes up/down.
    – TheCompWiz
    Jul 7, 2011 at 21:59
  • 1
    +1 to "let your routers route, use a first-hop redundancy protocol for your servers to always get to a gateway". Keep OSPF off of the servers. Jul 7, 2011 at 22:16
1

Generally, having servers participate in routing protocols is problematic; you're likely better off having them point at a floating virtual IP from your pair of BSD routers. It may also be possible to have your routers use a virtual IP as the origin for the BGP sessions, if your handoff is Ethernet or similar (where it's possible to do IP failover) and not an actual point-to-point interface (T1,etc).

Typically you want to have your BGP speakers all speaking to each other (or a route reflector, if your topology is big enough) as well as your upstreams, so everyone's on the same page, particularly when you're taking full tables from everyone. (Obviously you need to do some filtering, so as not to re-advertize routes from one provider to the other.)

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .