OSPF vs Static Routes for Multi Site Networks
Static routes are fine until the second path, the first dead neighbor, or the twelfth site. A practical guide to choosing OSPF or static routing per site, with configs, cost math, and a cutover procedure.

On this page
You can run a multi site network on static routes for years and never notice the cost. Then a second circuit arrives, a provider starts dropping packets without dropping the link, and the twelfth site needs a prefix that has to be reachable from every other site. The static table that was two lines per branch becomes a matrix that nobody wants to touch on a Friday afternoon.
The choice between OSPF and static routing is not a maturity test. It is a question about the shape of the network: how many paths exist between sites, how many prefixes have to be carried, how often the topology changes, and how much the team needs the network to discover a failure on its own. This guide works through those questions with real configurations, cost math, and a cutover procedure that can be reversed.
At a Glance: Static Routes vs OSPF vs iBGP
| Static routes | OSPF | iBGP | |
|---|---|---|---|
| Control plane | None, entries are configured by hand | Link state database and SPF | Path vector with policy |
| Best fit | Single path stub sites and simple hubs | Any to any sites with multiple paths | Internet edge and routing policy |
| Convergence | Interface down, or a tracked object | Hello and dead timers, or BFD | BGP hold timer, or BFD |
| Path selection | Fixed next hop and distance | Interface cost | Local preference, AS path, MED |
| Summarization | Manual aggregate with a next hop | Area range at an ABR | Aggregate with policy |
| Failure visibility | None beyond the tracked object | Neighbor state and LSA changes | Peer state and prefix counts |
| Operational load | One line per prefix and next hop | Areas, costs, and LSDB size | Peers, filters, and policy |
| Main risk | A blackhole nobody withdraws | A poorly planned area or cost map | A route leak or a filtered prefix |
Start With the Shape of the Network
The first useful question is how many paths exist between any two sites. A branch with one circuit and one exit has no path decision to make, so a static default route toward the hub is both accurate and cheap. A site with two circuits, or two sites that need to reach each other without transiting the hub, has a path decision that changes when a link fails. That is the point where a routing protocol starts earning its keep.
The second question is how many prefixes each site carries. Static routing scales with the product of sites and prefixes because every remote prefix needs an entry somewhere. OSPF scales with the number of links and the size of the link state database, and a new prefix inside an existing area is learned without a configuration change on the remote routers.
The third question is how often the topology changes. A network that adds a site every two years can absorb a manual change. A network that adds a VLAN or a subnet every week cannot, because each addition becomes a ticket that touches every router that needs to reach it.
Where Static Routing Is the Right Answer
Static routing is underrated in the places where it fits. A branch office with one circuit, one firewall, and a single summary prefix toward the hub does not need an adjacency, a link state database, or an area. It needs two facts: this is the local subnet, and everything else goes to the hub. Static routes also make a deliberate exception visible in the configuration, which is useful when a prefix must follow a specific path for policy or testing.
On Cisco IOS, the hub side of a small hub and spoke network can be as short as this.
ip route 10.20.0.0 255.255.255.0 192.0.2.1
ip route 10.21.0.0 255.255.255.0 192.0.2.1
ip route 10.22.0.0 255.255.255.0 192.0.2.1
The weakness appears when the next hop is reachable at Layer 2 but no longer forwards traffic. A static route does not notice a provider outage that leaves the physical interface up. The fix is to track something that reflects the path, not just the link.
ip sla 10
icmp-echo 10.20.0.1 source-interface GigabitEthernet0/0
frequency 5
ip sla schedule 10 life forever start-time now
track 10 ip sla 10 reachability
ip route 10.20.0.0 255.255.255.0 192.0.2.1 250 track 10
The static route now carries administrative distance 250, which makes it conditional: it is installed only while the probe succeeds, and it loses to any route with a lower distance. When the probe fails, the tracked object goes down and the route is withdrawn. If a primary route already exists, the same pattern makes the tracked route a floating backup. The probe target should be beyond the circuit, not the local provider address, or you will track the wrong failure.
Use static routing when all of these are true: one path per site, a prefix count small enough to read, a change rate your team can handle, and no requirement to share paths between branches. If any of those is false, the next section is the reason.
Where Static Routing Breaks
The arithmetic is the first warning. If every site has four prefixes and every site needs a route to every other site, the number of static entries grows with the square of the site count. A hub and spoke design reduces this, because each spoke only needs a default and the hub holds one entry per remote prefix, but the hub then becomes both the forwarding bottleneck and the place where every change lands.
| Sites | Prefixes per site | Full mesh static entries | Hub entries in a hub and spoke |
|---|---|---|---|
| 4 | 4 | 48 | 12 |
| 8 | 4 | 224 | 28 |
| 12 | 4 | 528 | 44 |
| 16 | 4 | 960 | 60 |
The second warning is failure behavior. A static route is a claim, not a measurement. When the circuit degrades but stays up, the route remains installed and traffic keeps flowing into a path that is dropping packets. A routing protocol at least has a neighbor that can time out, and with BFD it can detect a path failure in under a second. A static route needs a tracked probe to approximate the same behavior, and the probe has to be designed and maintained.
The third warning is change cost. Adding a VLAN to one site is a local change under OSPF. Under static routing it is a change on every router that needs to reach that VLAN, and the chance of missing one grows with every site. That is how networks end up with a static route that points to a subnet nobody has used since the last renumbering.
OSPF in One Page
OSPF is a link state protocol. Each router describes its own links in a Type 1 router LSA, floods that information within an area, and runs the shortest path first algorithm over the resulting database. The database is the same on every router in an area, which is why OSPF can converge without a central hub and why a failed link is visible to every router that was using it.
The cost of an interface is the reference bandwidth divided by the interface bandwidth, rounded up to an integer. The default reference is 100 Mbps, which makes every link at 100 Mbps or faster cost 1. That is the single most common OSPF design mistake in a modern network, because a 1G backup path and a 10G primary path look equal.
| Interface | Cost at 100 Mbps reference | Cost at 100 Gbps reference |
|---|---|---|
| 100 Mbps | 1 | 1000 |
| 1 Gbps | 1 | 100 |
| 10 Gbps | 1 | 10 |
| 100 Gbps | 1 | 1 |
Set the reference on every router to a value above your fastest link. If 400G is on the roadmap, use 400G or higher now, because changing it later changes every cost in the domain at once.
router ospf 1
router-id 10.255.0.11
auto-cost reference-bandwidth 100000
passive-interface default
no passive-interface GigabitEthernet0/0
network 10.255.0.11 0.0.0.0 area 0
network 192.0.2.0 0.0.0.3 area 0
!
interface GigabitEthernet0/0
description WAN to site B
ip ospf network point-to-point
ip ospf cost 10
ip ospf bfd
The passive-interface default line is deliberate. It stops OSPF from forming adjacencies on user facing interfaces, and then you opt in the WAN links one at a time. The point to point network type removes the DR election on a link that only ever has two routers.
Junos expresses the same design under the protocol hierarchy.
set protocols ospf reference-bandwidth 100g
set protocols ospf area 0.0.0.0 interface lo0.0 passive
set protocols ospf area 0.0.0.0 interface ge-0/0/0.0 metric 10
set protocols ospf area 0.0.0.0 interface ge-0/0/0.0 bfd-liveness-detection minimum-interval 300
set protocols ospf area 0.0.0.0 interface ge-0/0/0.0 bfd-liveness-detection multiplier 3
Verification is the part that separates a working adjacency from a configured one. Start with the neighbor table.
show ip ospf neighbor
Neighbor ID Pri State Dead Time Address Interface
10.255.0.12 0 FULL/ - 00:00:38 192.0.2.2 GigabitEthernet0/0
FULL is the state that matters. INIT or EXSTART points to a hello parameter mismatch or an MTU mismatch, not a routing problem. On Junos the equivalent is show ospf neighbor, and show ospf interface detail shows the timers and the interface cost.
If the cost is 1 on a 10G link, the reference bandwidth was never changed. That is visible in show ip ospf interface before it becomes a path selection incident.
Design the Areas Before You Enable OSPF
A single area works in a lab and becomes uncomfortable at scale. The practical design is one area per site, with the WAN core in area 0. Each site router that connects to both its local area and the backbone becomes an Area Border Router, and the ABR is where summarization happens.
The reason to split areas is not mystery. It keeps the link state database small at each site, contains the scope of an SPF run, and gives you a boundary where prefixes can be summarized. The reason not to overdo it is that every area needs at least one ABR, and a badly placed ABR can turn a simple failure into a partition.
On the ABR, an area range summarizes the intra area prefixes of the source area into a single Type 3 LSA. The command names the area the prefixes come from, not the area they are advertised into.
router ospf 1
area 10 range 10.20.0.0 255.255.0.0
area 20 range 10.21.0.0 255.255.0.0
area 30 range 10.22.0.0 255.255.0.0
Junos uses the same logic with a prefix instead of a mask.
set protocols ospf area 0.0.0.10 area-range 10.20.0.0/16
set protocols ospf area 0.0.0.20 area-range 10.21.0.0/16
set protocols ospf area 0.0.0.30 area-range 10.22.0.0/16
The ABR installs a discard route to Null0 for the summary by default, which prevents a loop when traffic arrives for a subnet inside the range that no longer exists. Do not remove that behavior unless you have a specific reason and a plan for the traffic that would otherwise follow a default route back into the network.
Branch areas can also be stub or NSSA. A stub area blocks external LSAs and relies on a default route from the ABR, which is appropriate when the branch has no local redistribution. An NSSA allows local external prefixes to be carried as Type 7 LSAs while still blocking Type 5 from the rest of the domain. A totally stubby area goes further and blocks inter area prefixes as well, leaving the default.
| Area type | Cisco config | What it blocks | When to use |
|---|---|---|---|
| Normal | default | Nothing | Sites that originate external prefixes |
| Stub | area 10 stub |
Type 5 external LSAs | Simple branch with no local redistribution |
| Totally stubby | area 10 stub no-summary |
Type 3 and Type 5, except a default | Small branch with few local prefixes |
| NSSA | area 10 nssa |
Type 5, allows Type 7 | Branch with a local firewall or redistributed prefix |
| Totally NSSA | area 10 nssa no-summary |
Type 3 and Type 5, allows Type 7 | Branch with local external routes and a small table |
Make Failure Detection Faster Than the Circuit
OSPF timers are chosen for stability, not for sub second failover. The defaults are fine on a LAN and too slow on a WAN that can fail in milliseconds.
| Network type | Hello | Dead |
|---|---|---|
| Broadcast | 10 s | 40 s |
| Point to point | 10 s | 40 s |
| NBMA | 30 s | 120 s |
| Point to multipoint | 30 s | 120 s |
Lowering the hello interval works, but it costs CPU and it makes the adjacency sensitive to loss. A better answer on links that can drop traffic without dropping the interface is BFD, which sends small packets at a negotiated interval and tears down the adjacency when they stop.
bfd interval 300 min_rx 300 multiplier 3
!
interface GigabitEthernet0/0
ip ospf bfd
That configuration detects a failure in roughly 900 ms, which is faster than most providers can reroute and fast enough to keep a voice call from noticing. The exact values should match what the platform and the circuit can support. Aggressive timers on a saturated link cause false positives, and a false positive is an outage you created.
Cisco also supports a minimal dead interval that derives the hello from the dead timer.
interface GigabitEthernet0/0
ip ospf dead-interval minimal hello-multiplier 3
That produces a hello every 333 ms and a dead interval of one second. It is useful on a point to point link between two devices you control. It is a poor default on a shared or congested path, and BFD is usually the cleaner tool because it is independent of the routing protocol.
Do not forget SPF and LSA throttling. A flapping link can cause more CPU load than the failure itself.
router ospf 1
timers throttle spf 50 200 5000
timers throttle lsa all 0 100 5000
The exact numbers depend on the platform. The principle is that the first SPF run should be fast, the hold time should back off, and the maximum wait should prevent a permanent storm.
A Cutover You Can Back Out Of
Moving from static to OSPF is safe if you treat it as an overlay first. The static routes keep winning on administrative distance while you prove the OSPF topology is correct, and you remove them only after the routing table agrees.
- Record every static route, its next hop, and the site prefix it serves. This is the baseline you will compare against.
- Assign a router ID per router, usually a loopback address, and decide area membership before any configuration is applied.
- Enable OSPF on the WAN interfaces only. Leave user facing interfaces passive so no adjacency forms where it should not.
- Set the reference bandwidth and interface costs on every router. Do this before the first adjacency comes up, or the first paths will be selected with the wrong costs.
- Confirm every neighbor reaches
FULLand that the LSDB contains the expected number of routers and prefixes. - Compare the OSPF route for each site prefix with the static route it will replace. They should point at the same next hop while both exist.
- Remove the static routes one site at a time, least critical first, and check the routing table, interface counters, and logs after each removal.
- Keep the static configuration in a rollback file. Restoring a static route is immediate because distance 1 beats OSPF distance 110.
The comparison is the step people skip. On Cisco, show ip route 10.20.0.0 255.255.0.0 tells you which protocol installed the route and which next hop it chose. While the static entry exists, the output says Known via "static", distance 1. After the static entry is removed, the same command should show ospf 1 with the expected metric and next hop. If it shows a different next hop, stop and find out why before you remove the next site. On Junos, show route 10.20.0.0/16 and show route protocol ospf give the same evidence. The route preference values differ from Cisco, so read the output rather than assuming a number.
Record the Routing Plan
The configuration is not the plan. The plan is the set of decisions that explain why the configuration looks the way it does: which router is the ABR for each area, which circuit sets each cost, which prefixes are summarized, and which static routes are deliberate exceptions.
The circuit inventory is the input to the cost table. If the inventory says a circuit is 1 Gbps and the OSPF cost assumes 10 Gbps, the routing plan and the physical network have already diverged. Pull the circuit list and compare it with the interface costs.
curl --fail --silent --show-error -H "Authorization: Bearer ${OBELINF_API_KEY}" \
"https://api.obelinf.com/${OBELINF_ORG}/circuits?all=true" \
| jq -r '.circuits[] | [.name, (.site_a_name // ""), (.site_b_name // ""), (.bandwidth // 0)] | @tsv'
wan-nyc-dc1 NYC DC1 1000
wan-nyc-dc1-backup NYC DC1 500
wan-dc1-lon DC1 LON 10000
The bandwidth field is in Mbps. A 1 Gbps circuit appears as 1000, so the same number can be compared with the interface bandwidth used in the cost calculation without a unit conversion. Record the OSPF cost next to the circuit and the mismatch becomes visible in one row.
Obelinf keeps sites, circuits, and device records in the same inventory, which makes it easier to review the area map and the cost table together. The multi site network use case describes how those records fit into a site by site operating model.
The practical next step is small. Pick one site, export its static routes, and build the OSPF configuration without removing anything. Compare the two routing tables for a week. If they agree, you have a cutover plan. If they do not, you have found the prefix nobody documented, which is exactly the problem the exercise was meant to surface.
Frequently Asked Questions
When should a multi site network use static routes instead of OSPF?
Does OSPF work over MPLS and internet VPN circuits?
Why do a 1 Gbps and a 10 Gbps link have the same OSPF cost?
What is the difference between an OSPF area range and a static summary route?
Can I run OSPF and static routes on the same router?
How do I migrate from static routes to OSPF without an outage?
How should I document an OSPF routing plan?
Stop reaching for a spreadsheet
Obelinf keeps every subnet, device, circuit, and rack in one live source of truth, with audit logs and a topology view. Free for personal use.
Related Articles

SNMP Traps vs Polling vs Streaming Telemetry
Polling, traps, and model driven telemetry describe the same switch but prove different things. Configure all three, then read each one for what it can actually confirm.
Read more
SNMP Explained: OIDs, Polling, and Real Data
SNMP is a database you query, not a monitoring system. Learn the OIDs that matter, why 32 bit counters lie, and how to turn a poll into a bandwidth number you can defend.
Read more
VRRP vs HSRP vs GLBP: First Hop Redundancy
Every host has one default gateway, and that gateway is a single point of failure until you make it virtual. Compare VRRP, HSRP, and GLBP with real configs, timers, and failure checks.
Read more