In order to provide augmented reality services, vast amounts of various forms of information are required, which can then be dispersed on multiple servers on the Internet. Thus, when a server does not have proper information for a certain request from a user device, it is better to pass the request to another server with the penalty of a longer response time. This paper proposes a method to reduce the response time by predicting the server that is most suitable for upcoming requests based on the locality of service requests from device using an augmented reality service. This is done by recording the servers that processed the requests, analyzing previous records to select the new server, and reestablishing the connection to this server to process the upcoming requests. Performance models of the proposed method are described and are evaluated in a simulation.