The problem now is how to measure the distance between machines in the Internet. Geographical distance is not a good enough metric because it completely ignores the topology of the Internet, which plays a key role in determining the actual connection speed.
A more promising solution is to use routing data to discover the path followed by the data sent to the client by the service. In this section we introduce a so-called ``AS-path length'' Internet distance metric. We then use it to develop a proximity-based policy that selects the replicas closest to the client. To understand better how the metric works, let us first explain some basics of routing in the Internet.