Workload dispatch planning for real-time fingerprint authentication on a sensor-client-server model

In this paper, we analyzed the collective performance of the task assignment of the fingerprint authentication on the sensor-client-server model. We first estimated the performance of primitive operations on the sensor, the client, and the server, respectively. Then, based on these primitive performance results, the workload of each scenario of the task assignment was applied to the M/D/1 queueing model representing the sensor-client-server model, and the collective performance of each scenario was analyzed quantitatively. The modeling results showed that extracting features and matching on the server could provide both fast response time and secure authentication with small numbers of clients. As the number of clients was increased, however, a bottleneck was observed on the server. The bottleneck can be eliminated by either performing the feature extraction on the client or increasing the number of processors in the server.