In order to direct content requests to an appropriate content serving site in
a computer network, a phased learning approach is utilized to ensure that, as best
as possible, the request is made to content serving site with the shortest delay.
In a setup phase, an indirect path return geographic sever load balancer times
sends transit time requests to all of the individual content serving sites so that
the transit requests all arrive at the content serving sites at the same time.
Therefore, when the requesting fixed location receives communications from the
content serving sites, it can easily tell which content serving site has the least
delay by an established race condition. The winner of the race may then be relayed
to the indirect path return geographic server load balancer for later usage. In
an execution mode, only the m fastest content serving sites and n other sites (used
to test random and new sites) are sent a transit time request when a DNS request
arrives from the requesting fixed location. The particular m fastest content serving
sites and n other sites may be dynamically updated so as to ensure the most reliable
directing of requests. This solution provides a very efficient and effective means
by which to determining closest content serving sites while keeping load balancer-created
traffic at a minimum.
