OSPF for ISPs

Your ISP's IGP is what holds everything together: BGP, MPLS, management. Badly designed OSPF turns into adjacency flap loops and instability across the whole backbone.

What OSPF is

OSPF is a link-state Interior Gateway Protocol (IGP) that computes the shortest path between routers inside the same AS. Unlike BGP (between ASes), OSPF resolves internal topology: how a packet gets from PoP X to PoP Y inside your network.

It's the de facto standard in mid-sized Brazilian ISP backbones. IS-IS is an alternative for very large networks — but OSPF covers 95% of cases.

What RASYS does with OSPF

  • Area design — well-sized area 0 (backbone), stub/totally-stub areas where they make sense to reduce LSA flooding.
  • Metrics and cost — manual cost to force preferred path (fiber link vs. radio). Avoid OSPF preferring sub-optimal route from auto-bandwidth bugs.
  • BFD (Bidirectional Forwarding Detection) — adjacency failure detection in milliseconds instead of seconds. Sub-second failover convergence.
  • OSPFv3 for IPv6 — dual operation with OSPFv2, no doubled overhead.
  • BGP redistribution — which internal routes should appear in iBGP/eBGP and vice-versa, with filters to avoid import-export loops.
  • Authentication — MD5/SHA between neighbors to prevent LSA injection from unauthorized equipment.
  • Troubleshooting — adjacencies stuck in INIT/EXSTART/LOADING, LSA debug, Graceful Restart during maintenance.

Equipment we work with

Huawei NE/AR series, Juniper MX/QFX, Cisco IOS/IOS XR/NX-OS, MikroTik RouterOS, FRR/Quagga on Linux.

When it makes sense to talk to us

You're growing and area 0 is huge; convergence is slow and customers complain; adjacencies keep dropping with no clear reason; want IPv6 in IGP without touching IPv4; consolidating two networks after acquisition.

Talk to us. See also: BGP, MPLS, OSPF in glossary.

FREQUENTLY ASKED QUESTIONS

How many OSPF areas does a regional ISP need?

Up to 50 routers: a single area (backbone area 0.0.0.0) works fine. 50-200 routers: 2-4 areas with a well-sized ABR. Beyond that, or with multiple PoPs, consider IS-IS, which scales better to hundreds of nodes. More areas than necessary adds complexity without benefit.

Does BFD alongside OSPF really speed up convergence?

Yes, drastically. OSPF default detects failure in 40s (hello 10s x dead-interval 4). With BFD at 100ms x 3 multiplier, detection drops to 300ms. Caution: overly aggressive BFD on an unstable link causes flap. We recommend 300ms x 3 (900ms) for edge, 100ms x 3 for optical backbone.

OSPF and BGP together — should you redistribute?

Do not redistribute BGP routes into OSPF (it explodes the table and creates potential loops). Redistribute OSPF/connected into BGP via network statement or redistribute connected with a strict filter. For own prefixes announced in BGP, use a network statement in BGP that mirrors what is in the IGP/connected.

Can OSPF on MikroTik handle a mid-size ISP?

Yes. RouterOS runs OSPFv2/v3 stably up to hundreds of adjacencies. The practical limit is CPU under heavy flap. For large BGP edge, MikroTik OSPF holds fine. For backbone with 500+ LSDB routes and multiple areas, carrier-grade equipment (Huawei NE40+, Juniper MX) scales better.