In Internet routing between autonomous systems which are interconnected in multiple locations, hot-potato routing is the practice of passing traffic off to another autonomous system as quickly as possible, thus using their network for wide-area transit. Cold-potato routing is the opposite, where the originating autonomous system internally forwards the packet until it is as near to the destination as possible.[1] [2] [3]
Hot-potato routing (or "closest exit routing")[2] is the normal behavior generally employed by most ISPs.[1] Like a hot potato in the hand,[2] the source of the packet tries to hand it off as quickly as possible in order to minimize the burden on its network.[1]
Cold-potato routing (or "best exit routing")[2] on the other hand, requires more work from the source network, but keeps traffic under its control for longer, allowing it to offer a higher end-to-end quality of service to its users.[1] It is prone to misconfiguration as well as poor coordination between two networks, which can result in unnecessarily circuitous paths.[1] NSFNET used cold-potato routing in the 90s.[2]
When a transit network with a hot-potato policy peers with a transit network employing cold-potato routing, traffic ratios between the two networks tend to be symmetric.[2]
Routing behavior can be influenced using two BGP "knobs": multi-exit discriminator (MED) and local preference.[1] In hot-potato routing, the MED attached to incoming -learned routes is discarded,[2] and the IGP cost is used instead.[3] In cold-potato routing, MED[2] or BGP communities are used to signal the cost of the route, which influences local preference.[3]