Looking for a sanity check on an OSPF configuration

[culor]

Hello

Been a while since I've done this and am hoping to get a sanity check before I get too deep. I'm inheriting a lot of strangeness and am trying to sort things out as I go; please excuse the mess. This is the situation:

Site A has two Catalyst 9500s (site-a1/site-a2). Site B has two Dell S5248F (site-b1/site-b2). We have a pair of singlemode dark fiber between sites with bidirectional SFPs, terminating in 25g SFPs (site-a1:tw1/0/29 > site-b1:eth1/1/48 and site-a2:tw1/0/29 > site-b2:eth1/1/48). The 9500s are stacked, 5248s are VLT'ed. I'd like to just address each interface separately, configure OSPF, and balance between the two fibers that way. That should allow me to keep both fibers active as well as allow me to fail between them. 

Any reason not to do this? I don't really see a better way to handle things, but I'm also sure I'm not the only one that has encountered this and someone smarter than me might be listening. We're not currently doing any other dynamic routing, although that will hopefully change soon. Am I on the right track?

[M02@rt37]

Hello @culor 

Assuming you're doing point-to-point L3 links directly between sites, you can configure them with /30 or /31 subnets and assign OSPF on each.

--Use OSPF network type point-to-point on each interface (for speed convergence and avoid DR/BDR election).

As long as both links are up and costs are equal, OSPF will install equal cost routes in the routing table.

[MHM Cisco World]

Using ospf to do some load balance is hard and not work always' 

Why you not try BGP??

MHM

[paul driver]

Hello

---

@culor wrote:
I'd like to just address each interface separately, configure OSPF, and balance between the two fibers that way. That should allow me to keep both fibers active as well as allow me to fail between them. 

---

Although you mention you would like to address each physical interface separately, you could aggregate them into L3 port-channels ( 95ks look like they do support them) this way the LB can be accomplished over a single ospf adjacency 

[Joseph W. Doherty]

Any reason not to do this?

It's been suggested that OSPF is hard and may not always work.  As least on Cisco platforms, if you have equal cost paths, to the destination, it's "automatic".  As to it doesn't always work, that's possible due to the traffic mix, for example if most the traffic is going to the same destination.  At least on Cisco, usually most (static and dynamic) routing can work the same, although BGP, possibility, is the most difficult, starting, by default, it only wants to use a single path.  It's usually not recommended as an IGP.  M02@rt37 describes how it's usually done.

As an alternative, possibility LACP bundle (what @paul driver also is suggesting) might be possible between the two hardware vendors.

At least on the Cisco hardware, you often have options how traffic is to be distributed across LACP multiple links.  Also, on the Cisco side, at least, you can do L2 and/or L3 across such a link.

[rkthapa9]

Hi @culor 

As you mentioned wanting to handle each physical interface individually, it may be more efficient to aggregate them into Layer 3 port-channels (which, it appears, the 95ks do support). This allows for load balancing over a single OSPF adjacency.

Some have suggested that OSPF can be unreliable or complicated, but on Cisco platforms, equal-cost multipath (ECMP) is handled automatically when equal-cost paths to a destination exist. Problems can still arise depending on traffic patterns—for instance, if most traffic is destined for the same endpoint, it might not distribute well.

Generally, both static and dynamic routing behave similarly on Cisco platforms, though BGP is a special case. By default, BGP prefers a single best path, and enabling multipath support requires additional configuration. It's also worth noting that BGP is typically not used as an interior gateway protocol (IGP).

As an alternative, LACP bundling between the two devices [as @paul driver  suggested] might be viable. Cisco platforms often allow you to configure how traffic is distributed across LACP links, supporting both Layer 2 and Layer 3 operation.

Thanks!