Interface Cost, Network Types and NBMA Multicast Problems

What it is

Interface cost decides which path OSPF prefers, and the network type decides how OSPF behaves on a link.

Each OSPF interface has a cost. The cost of a path is the sum of the costs of the outgoing interfaces along it, and the lowest sum wins.

The network type tells OSPF whether the link is shared, how neighbours are found and whether a DR is elected. Multicast is the root of most problems on non broadcast links.

How it works

You set a cost on each interface, and OSPF adds the costs along the path to compare routes.

Interface cost

In RouterOS the cost of an interface has a fixed default value. It is not calculated from the bandwidth, so a fast link and a slow link have the same cost until you change them.

/routing ospf interface add interface=ether1 cost=10
/routing ospf interface add interface=ether2 cost=100

To make OSPF prefer a fast link, give it a lower cost than the slow one. The cost has to be set by hand on every interface of a path, in a consistent scheme.

If two paths have the same total cost, OSPF can use both, as equal cost multi path.

Network types

The network type controls how OSPF treats the interface.

In RouterOS v6 the setting is network-type, with the values broadcast, nbma, point-to-point, ptmp, ptmp-broadcast and default. With default, the router picks a type from the interface.

/routing ospf interface add interface=ether1 network-type=broadcast
/routing ospf interface add interface=pptp-out1 network-type=point-to-point

In RouterOS v7 the equivalent setting is on the interface template, and it is called type.

Broadcast networks

On Ethernet, OSPF finds neighbours with multicast Hellos and elects a DR and BDR. This is the default behaviour and needs no special setup.

Point to point networks

On a link with exactly two routers, no DR is needed. Adjacencies form directly. This is the best choice for tunnels and for links that carry only two routers.

NBMA and multicast

NBMA stands for non broadcast multi access. Many routers share the network, but the network cannot deliver broadcast or multicast.

OSPF depends on multicast Hellos to find neighbours. On an NBMA network, the Hellos never reach the other routers, so adjacencies would not form.

The fix is to give the router a list of neighbours, and OSPF then sends unicast Hellos to them.

/routing ospf interface add interface=ether1 network-type=nbma
/routing ospf nbma-neighbor add address=10.0.0.2 priority=1

Problems with NBMA

NBMA still elects a DR, and the DR must reach all other routers directly. In a hub and spoke layout, the spokes cannot reach each other, so the hub must be the DR.

Set the priority of every spoke to 0 so it is never elected, and give the hub the highest priority. If a spoke becomes the DR, the other spokes cannot form adjacencies with it.

A second problem is the static list. Every router needs the right neighbours configured by hand, which does not scale.

Point to multipoint is a common alternative. It treats the network as a set of point to point links, uses multicast where it is available, and does not need a DR. It avoids the DR placement problem, at the cost of more LSA detail.

Verifying

/routing ospf interface print
/routing ospf neighbor print