New interns and externs achieve success with SIParCS 2014

By Brian Bevirt
10/01/2014 - 12:00am

Four students from CISL's class of 2014 are making their mark in the research community during their first summer at NCAR. Sunni Ivey, of the Georgia Institute of Technology, earned an Alfred P. Sloan scholarship to support her studies in Environmental Engineering. Two externs, Justin Moore of Salish Kootenai College and Lauren Patterson of Hampton University, used inexpensive Raspberry Pi clusters to learn concepts in high performance computing in a new CISL program designed to appeal to non-traditional students. Both externs presented posters at the August 2014 Rocky Mountain High Performance Computing Symposium. And Ye Feng of the University of Wyoming won best poster there for her work on accelerating a part of the Data Assimilation Research Testbed (DART) on Graphics Processing Units (GPUs). The three posters were all based on research conducted over 11 weeks in the summer.

Sunni IveySunni Ivey presenting the results of her SIParCS research project, Visualization of Air Quality Data in VAPOR, at the SIParCS symposium on 31 July 2014.

Sunni Ivey’s Alfred P. Sloan UCEM Scholarship will facilitate and complement her studies in the Environmental Engineering program at the Georgia Institute of Technology University Center of Exemplary Mentoring (UCEM). The funds will be used to cover personal and professional development activities that will accelerate and catalyze her success in graduate school. She also received Sloan Foundation support to attend the annual Institute on Teaching and Mentoring twice over the course of her graduate school career. The purpose of the Institute is to provide scholars with the skills necessary to succeed in graduate study and to prepare them for success as faculty members at colleges and universities. This four-day conference has become the largest gathering of minority doctoral scholars in the country. Sunni earned this honor after completing two undergraduate degrees, a Master’s degree, four teaching assignments, three internships, and having begun work on her Ph.D.

Three CISL students presented posters at the August 2014 conference organized by the Rocky Mountain Advanced Computing Consortium (RMACC). Held August 12-13 on the Boulder campus of the University of Colorado, the Rocky Mountain High Performance Computing Symposium focused on the continuing growth and future of high-performance computing in the Rocky Mountain region. Recognized as one of the top regional events on high-performance computing, the fourth annual symposium drew participants including designers and users of high-performance computing systems from universities, government laboratories, and industry from Colorado, Wyoming, Idaho, New Mexico, Utah, and Montana. The symposium featured nationally recognized speakers, group breakout sessions, a variety of tutorials covering many aspects of high-performance computing system design and use, and a student research poster competition. The winner of the poster presentation competition was awarded a trip to the Supercomputing 2014 conference in New Orleans.

Justine MooreJustin Moore presenting the results of his SIParCS research project, Performance Benchmarking a Raspberry Pi Cluster, at the SIParCS symposium on 1 August 2014.

Justin Moore’s research involved benchmarking a four-processor Raspberry Pi cluster with standard techniques used to evaluate and rank supercomputer performance internationally. Specifically, Justin optimized the parallel performance of the numerically intensive matrix-matrix multiply algorithm, comparing the performance of parallel versions built from basic linear algebra subprograms  (BLAS), machine optimized ATLAS libraries, and a hand-coded method for multiplying matrices of his own design. Justin went on to compute derived metrics of efficiency such as price- and power-performance, and compare his Raspberry Pi cluster with NCAR’ Yellowstone supercomputer in Wyoming using these.

Lauren PattersonLauren Patterson presenting the results of her SIParCS research project, Raspberry Pi Hadoop Cluster, at the SIParCS symposium on 31 July 2014. 

Lauren Patterson’s research used the Raspberry Pi cluster to compare ways of analyzing supercomputer interconnect topology to understand and improve parallel job performance. Lauren wrote graph traversal scripts using both the Hadoop MapReduce and message passing paradigms that converted interconnect configuration files and a list of nodes used a parallel job into lists of network pathways taken between all possible pair of nodes.  She prepared a poster about this research after her SIParCS presentation on the topic, and just over a week later she gave a talk and presented her poster at the RMACC symposium. Along with her mentor, Amogh Simha, she joined Justin, Raghu Raj, and a new faculty member at the University of Colorado, professor Karina Hauser for another talk where they described their research experiences using Raspberry Pi components to study supercomputing performance issues.

Raghu Raj Prasanna Kumar, a student assistant in CISL’s Technology Development Division who worked with the students on their summer projects said, “Justin and Lauren took advantage of many networking and public speaking opportunities at the RMACC conference. They each gave talks about their work with a Raspberry Pi cluster they personally built, and they also presented at the poster session. They were excited to take part in the conference and to compete for a trip to SC14 in New Orleans.  Between the SIParCS program and the RMACC conference, Justin returned to Salish Kootenai College and learned that his summer experience helped him earn an offer for a computing internship at his college. He accepted the offer and is well prepared to begin his new research project.”

Ye FengYe Feng of the University of Wyoming presenting the results of her SIParCS research project, Evaluating Coprocessor Effectiveness for the Data Assimilation Research Testbed, at the SIParCS symposium on 31 July 2014. 

Ye Feng’s research explored ways to accelerate the performance of the Data Assimilation Research Testbed (DART) using graphics processor hardware (general-purpose graphics processing units – GPGPUs). GPU technology is becoming an increasingly popular way to accelerate application performance by harnessing the massive amounts of parallelism inherent in their architecture. DART is a community facility developed and maintained at NCAR that offers powerful, streamlined tools that help researchers assimilate observational data into geophysical models. Ye and her mentor, Helen Kershaw in IMAGe, combined the two concepts in her summer research by applying the GPGPU acceleration to a computationally intensive DART function that performs nearest neighbor lookups. By implementing a new algorithm better suited to GPGPU’s, she was able to obtain a 10.9 times speedup relative to the original code running on a conventional Xeon processor. Judges at the HPC Symposium in Boulder awarded her poster top honors, including an all-expenses-paid trip to New Orleans for the November SC14 conference.