A cooperative approach to interdomain traffic engineering

For performance or cost reasons, autonomous systems (AS) often need to control the flow of their incoming interdomain traffic. Controlling its incoming traffic is a difficult task since it often implies influencing ASes on the path. The current BGP-based techniques that an AS can use for this purpose are primitive. Moreover, their effect is often difficult to predict. In this paper, we propose to solve this problem by using Virtual Peerings. A Virtual Peering is an IP tunnel between a border router of a source AS and a border router of a destination AS. This tunnel is established upon request from the destination AS. These tunnels can be negotiated by using backward compatible modifications to the border gateway protocol (BGP). By using Virtual Peerings, the source and destination ASes can achieve various traffic engineering objectives such as traffic-balancing or reducing the latency. A key advantage of our solution is that it does not require cooperation of the intermediate ASes and that it can be incrementally deployed in today's Internet. We then show by simulations that in a load-balancing scenario, a multi-homed AS only needs to request a few dozens of Virtual Peerings to balance its incoming traffic.