BGP calculates the best path toward a destination. With multiple ISPs in a multi-homed architecture, you can use the ISPs relationship to different networks on the internet to your advantage when load-sharing traffic.
There are three options for receiving routes from the ISP.
- Default Route only – in this scenario, you receive a default 0.0.0.0/0 route from the ISP but nothing else. This is easy to manage but offers no significant ability for load-sharing.
- A Partial Internet Routing Table plus Default Route – in this scenario, the customer receives a partial routing table containing prefixes that are in close proximity to the ISP ASN, plus a default route to handle all the rest of the internet traffic. This allows the customer’s BGP process to calculate paths and obtain some measure of load-sharing.
- A Full Internet BGP Table– in this scenario, the customer will receive the full BGP table, which is a heavy resource burden on the device. The administrator must make sure the appliance has enough RAM to accommodate this information, or else an outage can ensue.
The partial and full routing tables enable more optimal routing and provide load-sharing, but require complex configurations to support it.
A full routing table consumes RAM for storing the routing table, and CPU for calculating the best path. In 2014, there were approximately 500k IPv4 prefixes; today the internet is approaching 768k prefixes.
You must always be careful to filter routes at your internet edge, or risk becoming a transit AS for other internet prefixes, which causes your corporate network to route traffic for the rest of the internet.
Stuck-in-Active: Journal of an IT-Network Administrator 