Hi,
When a client makes a DNS lookup using its local recursive DNS (LDNS), that LDNS is the device that makes the query to the GTM, so the GTM (with the exception of recent additions to the DNS protocol), has no way to know what the original client's IP address was, but in many cases, the IP address of the LDNS is somewhat related to the client's IP. They tend to be in the same country, or regional area, because an ISP wants their DNS servers to be as close as practical to the users to reduce DNS lookup time.
So the round trip from a data centre to an LDNS can be used as an approximation for the experience that the user would observe. To do this, you can set the preferred load balancing method for the GTM pool to one of the dynamic methods, for example, Round Trip Time, Hops, Packet Loss. Selecting one of these methods causes the GTM to send out probes to the LDNS servers, once we receive a request from them. Note that the Alternate load balacing method drop box does not contain any of the dynamic methods, as there has to be a non-dynamic method available for the first few request, since we won't have the LDNS probe data when the first request from that LDNS arrives at the GTM.
The configuration related to LDNS metric collection is under DNS / Settings / GSLB / Metrics Collection, and you can select the type of probe the GTM should send. By default, it's just a ping, but we also support a TCP connection, a UDP connection, and two types of DNS queries.
The use case here is that if you had two GTMs in two data centres, say East Coast and West Coast of the US, and the probe RTT for the East Coast GTM was 60ms, but the RTT for the same LDNS from the other DC was 260ms, then the GTM can infer that the client is much closer to the East Coast DC, and that the client that used that LDNS would likely be best served with a response from a device in the East Coast.
For more information about this, and more complete descriptions of the dynamic methods, please refer to this manual page