MRI: Acquisition of a Heterogeneous Computing Platform for Biometrics Research

Clarkson University, Potsdam NY

Investigators

Abstract

This project, acquiring a heterogeneous high-performance computing cluster, aims to support parallel processing research of biometrics and identification technology, as well as broad disciplines of engineering research. Operational capabilities of managing and analyzing large-scale biometric information in an effective and efficient manner constitutes a major challenge faced by researchers in advancing biometrics. Emerging computing elements, such as many-core processors and hardware coprocessors, play an essential role in achieving this goal. This project enables the proponents to investigate novel applications of emerging hardware technology to a problem of current national interest. The platform should achieve effectiveness with great performance from its heterogeneous architecture and efficiency with power-awareness and energy awareness. Both the biometrics and high-performance research computing community will gain from the heterogeneous high-performance computing platform that can employ various state-of-the-art parallel architectures for hardware acceleration of biometric applications. It can serve as a design reference for next-generation commercial and governmental biometric systems. Since this institution currently serves as the lead site of the Center for Identification Technology Research (CITeR), a multi-university NSF I/UCRC, the instrumentation serves as a great enabler in support of continued research efforts of its affiliates interests as these evolve towards more advanced research in high-performance computing aspects of biometrics. This instrumentation provides the capability for the researchers to contribute towards and advance the parallel processing of biometric applications on heterogeneous computing platforms. The system lends suitable capability for processing a wide range of biometrics applications beyond those currently available. Moreover, the equipment also supports efforts to compete for other competitive research. The cluster consist of Central Processing Units (CPUs), Graphics Processing Units (GPUs), Many-Integrated Core (MIC) co-processors, and Field-Programmable Gate Arrays (FPGAs), tightly integrated with a light field camera as a data-capturing front-end. The high-performance computing community acknowledges that with the transition from single-core processor to multi/ many-core processors, no one single processing element can achieve the best performance for biometrics applications (as well as other different applications) since often different parts of the program have different parallelism characteristics suitable for acceleration by different processing elements. Inherited in biometric applications a large degree of data parallelism exists that requires carefully mapping the different region of the biometric applications onto different hardware components and orchestrating them to function as whole, so as to produce results in an effective and efficient manner. So, in order to achieve the best performance, a combination of computing elements need to be used.