A carrier will normally have an interlink - this could be a dedicated port on the FB9000 or a VLAN perhaps via a suitable switch. In any case this is an interface in the configuration. Some carriers use a /30 IPv4 interlink per LNS, and some use a larger subnet covering several LNSs.

This interlink is usually used solely for the purpose of a BGP link to the carrier, and all other IPs used by the ISP or carrier are announced via that BGP connection. You may want to configure filters on the BGP connection to limit the prefixes accepted from the carrier or announced to the carrier.

An alternative approach is to configure the interlink interface on a separate routing table. The FB9000 can have separate routing tables which act as completely separate internets. Using a separate table means you do not have to worry about what prefixes are announced on the BGP link as they will only apply to that routing table.

Whilst we would recommend using a BGP connection like this, if the carrier does not handle BGP then you will need static routes. Using a separate routing table can make this much simpler as you can set a default route to the carrier gateway on the interlink subnet.

If using a separate routing table, you have to take care to correctly configure the routing table on the interface, BGP, RADIUS, L2TP and loopback configuration elements.

We recommend using RADIUS session steering with the carrier, if they support it. Session steering means that the carrier sends a RADIUS request to the ISP before connecting each session by L2TP. The reply steers the connection to a specific LNS. The connection details include the target (IP address) which will be one of the FireBrick's address, and a pre-agreed hostname which identifies the tunnel level connection, along with a secret to authentication the connection. Obviously these details have to match what the FireBrick is expecting in its L2TP configuration. Session steering gives a lot of control to the ISP, and is ideal if you operate bonding connections where multiple links need to use the same LNS.

The carrier will typically expect you to have two RADIUS endpoints to which they can send requests. One for master and one for backup. Whilst the FB9000 will answer RADIUS on any of its IP addresses, we know some carriers have issues using the interlink IP addresses. We recommend you create two additional loopback addresses for session steering RADIUS.

These addresses are configured as a BGP announced loopback address. You can use MEDs to steer which IP is on which LNSs. If you have more than two LNSs you can ensure that the same IPs are announced from more than one LNS, and let BGP decide which LNS gets the RADIUS requests. RADIUS is a simple question and answer protocol so it does not matter which LNS gets the request.

The session steering configuration (Platform RADIUS) is very flexible. We suggest you use the same configuration on each LNS so that the replies are consistent regardless of where the request goes. The reply says where to connect the session (as well as hostname and password to use). The reply can be a single LNS or can be more than one reply with a priority tagging if the carrier supports this. The FB9000 can pick an LNS randomly from a set, or pick one based on a hash of the username, part of the username, or circuit ID. This can be useful when multiple lines are in a bonded arrangement where using the same prefix on a username ensures all of the connections are steered to the same LNS for bonding.

If you have a lot of LNSs we recommend an N+1 arrangement. E.g. if you have 4 LNSs you may set your RADIUS session steering to reply with one of three active LNSs as the first choice (perhaps using a hash) and the 4th (backup) LNS as a second choice. This keeps one LNS as a hot standby for failure and also allows maintenance on it, such as s/w updates. You can then change which of the there LNSs are active and either wait for lines to move when they reconnect or clear lines on the new backup LNS. This makes it easy to do rolling upgrades on s/w or other maintenance.

Session steering also allows specific configurations to be based on username, and circuit and so on, so allowing different responses for different carriers and different end users to be customised if necessary. It is also possible to send a copy of the session steering RADIUS to your own RADIUS server for logging.

The FB9000 will accept L2TP connections on any of its IP addresses, but again we recommend allocating a loopback address or using the address from the LAN rather than the interlink address as we know some carriers cannot handle that.

Unlike RADIUS where any request can go to any LNS, the L2TP connections have a state, and so you will want the address to stay with the particular LNS. Do not have a loopback that floats between LNSs via BGP as this would mean existing connections trying to talk to another LNS. It is better for the a failure to cause a clean timeout on the failed LNS and reconnect (via RADIUS session steering) than to have an active session's traffic go to a different LNS where the session IDs could match an existing session (unlikely, but possible).

The L2TP connection is matched to an incoming L2TP configuration. The RADIUS session steering can be used to specify the hostname and password that is sent so that the correct incoming configuration can be matched. This can select the specific RADIUS servers to use at the ISP for authorising the connection (though typically a single set of RADIUS servers is used for all connections). It can also specify defaults for DNS, PPP endpoint addresses and so on.

Once the L2TP connection arrives you can use RADIUS in your own network to control the connection, accepting it or rejecting it, and defining IP addressing, DNS, traffic speeds, routing table, and much more. Appendix G provides details of the specific AVPs used with RADIUS for L2TP.

You would normally have more than one RADIUS server. You can set these in a priority order, for example a set of main servers and a set of backup. The FB9000 will find a config line for RADIUS based on the named RADIUS server in the L2TP incoming configuration, or pick any if this is not set. It checks these in order. Each RADIUS configuration can have multiple servers.

Having picked a set of RADIUS servers to try they are sorted based on their previous response time and reliability. The requests are then sent to servers in order, allowing enough time for a response based on previous performance. There are settings to fine tune these timings. Once a response is received then the L2TP connection can proceed.

The same process is followed for RADIUS accounting. Each config can say if it is used for authentication or accounting or both.