Connecting Multiple ISPs Using BGP

The generic term , multihomed , is often used to depict a host linked to two or more networks or having two or more network addresses. For instance, a server might be connected to a serial line and a LAN. For the purposes of Cisco routing, generating multiple connections between an autonomous system and two or more ISPs is referred to as multihoming . The technique of multihoming is often used to establish redundant connections between ASs and the Internet to increase the stability and reliability of the connection or to utilize preferable paths to particular destinations with which it often communicates. By generating links with more than one ISP, an organization can ensure that communications with Internet destinations are not broken if a particular ISP experiences a failure.

BGP Multihoming is technically not a configuration of multiple links to a single ISP, such as an ISDN fault-tolerant connection. Although multiple links to an ISP do help in the event of a link failure, they do not protect your organization against a communications failure at the ISP itself.

By establishing connections to more than one ISP, you can also improve performance between the AS and common network destinations. For example, if your two most common destinations used different ISPs and a separate link was established with each ISP, then certain routes could be established through each ISP between the AS and common destinations. This offers more possibilities for policy control than would be possible with only a single link to one service provider. For example, a company could use a particular provider for multicast conferencing services and instant messaging while using another provider for email services.

Three common ways exist to configure multiple connections to the ISPs. One method entails all ISPs passing only default routes to your AS. A second method is to have all ISPs pass default routes and selected routes to your AS (the selected routes, for example, of certain high-traffic customers). Third, all ISPs can pass all routes to your AS. The following sections look at each of the scenarios in detail.

All ISPs Pass Only Default Routes

If only default routes are passed from each ISP to the AS, for example, then memory and CPU utilization are held to a minimum. Your autonomous system passes all its routes to the multiple ISPs, who in turn apply policy and forward them on to other ASs as necessary. Inbound packets coming into the AS are managed by the ISPs and other ASs, whereas the outbound packets going to the ISPs are managed according to the metrics of the IGP, such as RIP, EIGRP, or OSPF, which are used to reach the default route configured internally at the local AS. The route to the ISP depends upon the metric of the routing protocol used as the IGP. At this point, you should be well aware of how to use the ip route command to configure a default route. However, here is a quick example of configuring interface Serial 0 as your default route:

 RouterA(config)# ip route 0.0.0.0 0.0.0.0 S0 

All ISPs Pass Default and Selected Routes

Establishing specific routes to common destinations demands additional router overhead (processor and memory) because both default routes and specific routes to other destinations are being processed . You want to choose only customers with whom you exchange a substantial amount of traffic to pass selected routes. The ISP that a router in your AS selects to reach a customer network is most likely the one with the shortest AS path value (however, this could be rigged with a Local-preference attribute, for example). The default route configured and used by the IGP will be used to get to all the other external networks. As in the previous scenario, the ISPs and other ASs control inbound packets coming into the AS.

All ISPs Pass All Routes

A multihomed AS can be configured to receive all routes from each ISP. Obviously, this significantly increases the overhead for each router in your AS because all external routes have to be processed. Your AS sends all its routes to each ISP, which in turn applies policy and forwards them when applicable to other ASs. The ISP that a router in your AS selects to reach an external network is most likely the one with the shortest AS path value, although attributes such as Weight and Local-preference can be used to influence this decision-making process.

Weight and Local-preference Revisited

Some of the most common methods for manipulating the path chosen to reach external networks are the Weight and Local-preference attributes. The following snippet shows the syntax used to designate a weight attribute:

 RouterA(config-router)#  neighbor  {ip-address  peer-group-name}  weight  weight 

The ip-address parameter is the IP address of a BGP peer, and peer-group-name is the name of the BGP peer group. Although either the ip-address or peer-group-name parameters can be used, neither are required. The weight keyword is, of course, the assigned weight value, which can be between 1 and 65535 (16-bit). The default weight value is 32768 for routes that originate from the local router. The default weight value for other routes is 0. The weight value influences only the local router.

You can modify the default local preference value from 100 to another number between 1 and 4,294,967,295, where the higher value is preferred. This value is local to the autonomous system and is utilized to influence routes that have equal weights. It influences other routers within the AS. The command syntax looks like this:

 RouterA(config-router)#  bgp default local-preference   value  

Although there are several other BGP commands that can be used to manipulate and influence router policy, the rest of these options typically utilize route maps. Route maps are explored in greater detail in Chapter 10, "Optimizing Routing Updates."