Disclosed are a global server load balancer (GSLB) apparatus and a method for dynamically controlling cache time-to-live (TTL) in the GSLB apparatus. The GSLB apparatus which performs load balancing among a plurality of local servers includes a domain name system (DNS) query processor that receives a query message from a DNS server, a server selector that selects a local server among the plurality of local servers based on the received query message, a load calculator that collects state information of the GSLB apparatus, and calculates a load of the GSLB apparatus from the collected state information, and a TTL controller that updates a TTL of the GSLB apparatus based on the calculated load.