K-Test: An Analytical Model for Measurement-Based Server Selection Based on Response Time

Different paradigms employ server selection to establish a fast connection to end users, thereby reducing the delay associated with exchanging data between mobile users and servers. To efficiently infer the statistics of the server response, we propose an analytical model for server selection based on
$K$
measured response times, namely, the K-Test method. First, the probability that the user device correctly selects the best server and the corresponding time taken in K-Test method are investigated. Then, we study the server switch when connection failure is encountered. The proposed analytical model for K-Test method considers the distribution of the response time of candidate servers and derives three metrics, namely the probability of selecting the best server, the expected latency of the server selection process, and the false connection failure detection probability. The analytical model is validated by the simulation. Numerical results show that both the value of
$K$
and the number of candidate servers significantly affect the selection probability. Also, it needs more test rounds to correctly find the best server when users experience a variation in the network response times.