Cost-performance analysis of cloud-based biometric systems.

摘要

Iris and other biometric-based identify systems are being developed today in large and larger contexts. Much of the past research has investigated the accuracy of these systems, yet deploying into the range of tens of millions of records is largely an unsolved problem. Researchers have adapted the Hadoop Map-Reduce engine on top of multi-machine clusters to perform these calculations. These systems can be very expensive, but by carefully examining the running characteristics, cost can be effectively managed as a function of desired speed and database size. Various hardware congurations are investigated as well as a discussion of optimal Hadoop cluster settings specically for the Iris Recognition problem. The results culminate in actionable guidelines for the constructions and management of a cost-effective Hadoop based biometric platform.