It would seem that information about the state of the network may provide a valuable means for determining the best POP to handle a client's connection. A readily available network metric is ping-time, a meausre of the round trip time for a packet from one host to another and back again. Unfortunately, ping-times are not a good measure of network performance. It is quite possible for a high latency link to be a high bandwidth link. Satellite and DSL are good examples of this. It is of note that much Internet traffic, including HTTP, SMTP, FTP and streaming media are bandwidth, rather than latency sensitive. The usefulness of ping-times is further compromised if asymetric routing is being used, as the ping-time will not provide information on any differences in the forward and return trip times.
The collection of ping-times is also problematic as having POPs ping a client is inherently slow. It is possible to cache the results but avoiding a delay when a client first connects is difficult. A ping may also be blocked by a packet filter for some reason. If ping packets are being blocked, at best the ping-time data will be unavailable and at worst the ping may time-out, adding to the delay in establishing a connection for the client.