Deflection routing
From Wikipedia, the free encyclopedia
Deflection routing, also known as hot potato routing, is a routing strategy for networks based on packet switching which can reduce the need of buffering packets. Every packet has preferred outputs along which it wants to leave the router, and when possible a packet is sent along one of these outputs. However, two or more packets may want to leave along the same output (which is referred to as a contention among packets), and then only one of the packets may be sent along the link, while the others are sent along available outputs, even though the other links are not preferred by the packets (because, for instance, those links do not yield shortest paths).
Depending on the rate of incoming packets and the capacity of the outgoing links, deflection routing can work without any packet buffering. Of course, it is always possible to simply drop packets in a network with a best effort delivery strategy.
In the jargon of routing technology, hot-potato routing is routing that forwards the packet towards the path with the lowest delay (as opposed, for example, to the more commonly used metric of smallest number of hops).
[edit] Use of the terms "hot-potato routing" and "cold-potato routing" in commercial networking
In commercial network routing between autonomous systems which are interconnected in multiple locations, hot-potato routing is the practice of passing traffic off to another AS as quickly as possible, thus using their network for wide-area transit. Cold-potato routing is the opposite, where the originating AS holds onto the packet until it is as near to the destination as possible.
Hot-potato routing is the normal behavior of most settlement-free peering agreements. Hot-potato routing has the effect that the network receiving the data bears the cost of carrying it between cities. When the traffic ratio (the ratio of traffic flowing in one direction to the traffic flowing in the other direction between peers) is reasonably even, this is considered fair, because the networks will share evenly in carrying traffic exchanged by their customers between cities.
The marginal cost of carrying traffic between cities depends on how the network has purchased those links; some networks own dark fiber, which can be upgraded by merely replacing the equipment on each end of the fiber, and possibly the amplifiers along the path between cities. In other cases, the network has an agreement with a telco that allows for a specific amount of bandwidth, and upgrading involves paying more money to the telco.
Cold-potato routing, on the other hand, is more expensive to do, but keeps the traffic under your control for longer, allowing operators of well-provisioned networks to offer a higher QoS to their customers. It can also be preferred when connecting to content providers; if content providers use hot-potato routing, they may escape from paying for the cost of links between cities.
Cold-potato routing is prone to misconfiguration as well as poor coordination between two networks. In such scenarios, packets can be routed further distances as well as allow another autonomous system to manipulate routing in your network for various purposes. Cold-potato routing requires a level of trust between two networks that either side will not attempt to "cheat" the other.
Some content networks favor the use of cold-potato routing (MED exchange/honoring) in order to deliver content from replicated server farms closer to the end-user.
The terms can also be used to describe the route announcement policy of a network: by choosing to announce their network at a large number of points at the periphery of another AS, a provider can pull incoming traffic onto their network as soon as possible, ensuring that the traffic stays on their network all the way to their customer's connection.
