A† Map-Reduce Paralleled/Distributed Computing Model for Analyzing

the Structure and† Properties of Materials


Ziwen Fu, Wenwu Chen and Paul Kienzle


††††† Map-Reduce is a programming model for massively paralleled/distributed computing. The reflectometry team at the NCNR has developed a fitting service using map reduce, allowing scientists to perform constrained global fits to multiple datasets using computer clusters.† As part of DANSE (Distributed Data Analysis for Neutron Scattering Experiments) the software will provide various models of interest to neutron scattering practitioners, but the framework is extensible so models from other domains can be added. The fitting engine is controlled via TCP/IP from a Python backend.† Users can drive the fit through scripts or use a traditional graphical user interface to build the models and fit them to data sets. †The software supports common desktop and server operating systems (Windows, Linux, Unix and OS X).


††††† This analysis method can simultaneously search the global best solutions for multiple models. For example, given a sample description (e.g., a multilayer system composed of Al-Gd-EuO-Cu-Cr-Si), we can model the expected neutron and X-ray reflection for the model as a function of the thickness of the constituent layers and match this with measured data.† Analyzed separately, the individual data sets cannot determine the depths of all the layers.† However, since the contrast between layers is different for neutron and X-ray reflections, we can find reasonable physical solutions by fitting both data sets simultaneously. †Using this technique we can more easily get a consistent understanding of our sample using information from many different types of measurements.


††††† We will show results of constrained searches using paralleled/distributed computing and simultaneous refinement for exchange-biased spin valves, systems which are important to the development of high density storage magnetic storage devices.



Author Information

            Name:                           Ziwen Fu

            Mentorís name:            Paul Kienzle

            Division:                       NIST Center for Neutron Research -610

            Laboratory:                   Materials Science and Engineering Laboratory

            Building:                       235

            Room:                           E109

            Mail Stop:                     8562

            Telephone #:                 301-975-4386

            Fax #:                            301-921-9847

            Email:                            fzw@nist.gov

            Sigma Xi:                      No

            Category:                       Materials Science