SIParCS interns and student assistants sizzle this summer

By Richard Loft
10/07/2015 - 12:00am

The weather may have been cooler and wetter than normal in Boulder this past summer, but the projects done by CISL's class of SIParCS interns and student assistants certainly sizzled.

The 2015 class of the Summer Internships in Parallel Computational Science (SIParCS) program was the second largest in its nine-year history, with 17 students participating. It was also the most diverse, with three interns from Minority-Serving Institutions (MSIs), five female interns, and eight interns from seven different EPSCoR states. EPSCoR states are defined as those determined by the government to be underserved by federal research and education funding, and are thus eligible to receive funds to advance their research infrastructure through the Experimental Program to Stimulate Competitive Research (EPSCoR).

Also varied were the types of projects students undertook with their mentors: they ranged from a team of four undergraduate students designing and building an end-to-end, cloud-based workflow for collecting, storing, and displaying real-time weather data on a website using low-cost Raspberry-Pi processors, to interns porting, benchmarking, and analyzing the performance of the WRF weather model. Others worked with graphics processing units (GPUs) to do raycasting of volumetric data (computing a 2D image from a 3D data set), simulate atmospheric equations, assimilate data into forecast models, or solve geostatistics problems, while others focused on enhancing the performance of climate models and climate data analysis workflows. Still others used high-performance tools to shed light on real-world geostatistics problems, such as understanding future changes in Rocky Mountain snow pack in response to projected climate change.

Sample template image
The SIParCS class of 2015 included 17 interns, plus one early-career visitor, Qian Zhang, who recently received her Ph.D. and was affiliated with the program. Pictured here are, left to right: Delilah Feng, Shreya Mittapalli, Mudambi Srivatsa, Whitney Nelson, Negin Sobhani, Vinay Ramakrishnaiah, Ian Bragg, Qian Zhang, Dongliang Chu, Sam Elliott, Colette Smirniotis, Gaston Seneza, Adnan Haider, Lee Richardson, Jenish Koirala, Albert Yau, Kyle Marcus, and Tuan Ta. SIParCS is an 11-week summer program run by CISL that provides undergraduate and graduate students with an opportunity work on research projects in the computational sciences that span computer science, applied mathematics, statistics, visualization, and software engineering. (Photo by Brian Bevirt.)

During the summer, SIParCS students had an opportunity for enrichment activities such as a technical writing seminar, high performance computing (HPC) training classes, and a field trip to the NCAR-Wyoming Supercomputing Center (NWSC) to see the Yellowstone supercomputer. At the end of the summer program, the interns presented their research results in NCAR’s main seminar room from 29–31 July.

In the end, the wet weather didn’t dampen students’ enthusiasm for their SIParCS experience. “After this internship, I’ll introduce SiParCS to my research group as a great place for a good internship,” Tuan Ta said, “I really appreciate your support!”

But for some, the summer fun was not over: some students opted to attend, with CISL/SIParCS support, a regional HPC conference to gain additional professional experiences.

The 2015 Rocky Mountain Advanced Computing Consortium (RMACC) High Performance Computing Symposium was held 13–15 August in the Wolf Law Building on the Boulder campus of the University of Colorado. Attendance at RMACC-2015 exceeded 300, a new record, and included over 100 students. The annual symposium is focused on the continuing growth and future of high-performance computing in the Rocky Mountain region, and is recognized as one of the top regional events in high-performance computing in the country. The fourth annual RMACC Symposium drew participants including vendors and operators 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, and a variety of tutorials covering many aspects of high-performance computing system design and use. RMACC-2015 also featured a student research poster competition, with winners receiving trips to attend the IEEE Supercomputing Conference in Austin, Texas this fall.

Students with a CISL connection also did very well in the RMACC-2015 poster competition, winning three of the four prizes awarded by RMACC judges. A record total of a seven CISL students competed. “The only prize we didn’t win was in computational biology, which is understandable given NCAR’s mission,” said a smiling Raj Kumar, a SIParCS mentor of one of the winners, “I am very proud of all of our students and student assistants!”

Negin Sobhani, from the University of Iowa, won for her research poster entitled “Performance Analysis, Profiling and Optimization of Weather Research and Forecasting (WRF) Model.” As a SIParCS intern, Negin worked over the summer with her mentors Davide DelVento (CISL User Services) and Dave Gill (of MMM) to examine at the performance of Advanced Research WRF (ARW) using the performance profiling tools Vtune Amplifier and Tau, to zero in on the model’s performance issues. ARW is a widely used, fully compressible, nonhydrostatic weather model with complex dynamics and physics solver components. Negin analyzed its performance and discovered computational hotspots in certain calculations related to the transport of trace gases by winds, called advection, in the ARW model, and proposed approaches for making these calculations more efficient.

Sample template image
Negin Sobhani (right), a SIParCS student from Iowa State, explains her research on profiling the WRF model to attendees of the 2015 RMACC Symposium in Boulder, Colorado. (Photo by Brian Bevirt.)

 

Sample template image
Tuan Ta, a SIParCS intern this past summer, presents his GPU-CPU performance portability results at the RMACC 2015 poster competition. (Photo by Brian Bevirt.)

Another winner was Tuan Ta, an undergraduate student from the University of Mississippi and SIParCS intern, whose summer research compared the optimal performance of advanced central processing units (CPUs) and graphics processor units (GPUs) on the shallow water equations. These equations are a simplified, two-dimensional representation of the non-linear equations that govern the motion of the atmosphere.

Working with mentors Raj Kumar of the Application Scalability and Performance Group and Natasha Flyer of IMAGe, Tuan wrote and optimized different versions of the finite difference radial basis function (RBF-FD) method of solving these equations on different computer platforms and also using a variety of parallel programming languages. Tuan’s work suggests that performance portability – the ability to move application code between CPUs and GPUs without losing performance – is possible, at least in some cases.

Sample template image
Melissa Rishel, a student at the University of Northern Colorado and a student assistant in CISL during the summer of 2015, speaks on her summer research that analyzed CISL supercomputer user survey results to estimate the scientific publications attributable to the Yellowstone supercomputer. (Photo by Brian Bevirt.)

The third CISL poster competition winner was Melissa Rishel, a graduate student starting her second year of studies in Applied Statistics and Research Methods at the University of Northern Colorado in Greeley. Melissa first visited NCAR with her son’s second grade class on a field trip in October 2013. After completing her second semester in graduate school, Melissa applied and was offered a summer position in the User Services Section working for Dave Hart and with Doug Nychka in IMAGe.

Her project applied statistical methods to assess how well the annual user survey captures the total number of peer-reviewed publications produced in the previous fiscal year by users of the Yellowstone supercomputer. Her work also sought to determine the error bounds that could be put on the estimated total number of publications.

Melissa was delighted to be a part of the NCAR community for the summer. “The research project truly complemented my studies as I hope to continue along the lines with data analysis, prediction modeling, and also data mining,” Melissa said.

Also participating in this year’s poster competition was SIParCS intern Vinay Ramakrishnaiah, a graduate student at the University of Wyoming, who worked with mentors Dorit Hammerling (in IMAGe) and Raj Kumar (ASAP) on a research project entitled “Profiler Guided Manual Optimization for Accelerating Cholesky Decomposition in the R Environment.” Vinay studied ways to accelerate one of the biggest computational challenges in geostatistics: the Kriging algorithm, which requires Maximum Likelihood Estimation, which in turn involves costly Cholesky decomposition operations, whose computational costs increase as the cube of the number observations. Vinay used the Matrix Algebra on GPU and Multicore Architectures (MAGMA) library to significantly accelerate Cholesky decompositions. Ye Feng, who goes by the name Delilah, also from the University of Wyoming, was a returning SIParCS intern from 2014. Delilah authored the research poster called “Evaluating Coprocessor Effectiveness for DART,” and her work on the Data Assimilation Research Testbed (DART) was a continuation of her prize-winning work from last year, which uses GPUs to accelerate a key, memory-bound portion of DART that performs linear regression of a state variable onto an observation and computes the state variable increments from the observation increments.
Sample template image
Priyanka Sanghavi (right) discusses her results on low-cost computer vision computations with an RMACC 2015 attendee. Priyanka is a Master’s degree student at CU Boulder and a CISL student assistant who helped several SIParCS interns this past summer. Priyanka did the research she presented at RMACC “on the side.” (Photo by Brian Bevirt.)

Finally, CISL student assistant Priyanka Sanghavi benchmarked computer vision kernels on the low-cost, NVIDIA Jetson TK-1 graphics processor to evaluate their utility in portable devices for the visually impaired. Rashmi Oak, another student assistant in CISL, used NVIDIA’s GPU-Direct communication library to enhance the performance of MAGMA linear algebra library functions when multiple GPU cards are available. Both Rashmi and Priyanka are Master’s students in the Electrical and Computer Engineering program at the University of Colorado, Boulder.

The SIParCS program provides a wide range of valuable work experience for high-performing undergraduate and graduate students. In return, these students perform useful work that directly advances the first two parts of CISL’s mission. This internship program therefore aligns with all three parts of CISL’s mission:

  • To substantially enhance the productivity of a growing community of researchers in the atmospheric and related sciences through advanced computing and data services.
  • To improve computational methods for current and future high performance computing systems.
  • To prepare future generations of scientists and engineers for current and future supercomputing challenges.

The ultimate goal of SIParCS is to address shortages of trained scientists and engineers capable of using and maintaining these high-end systems to achieve the goals of 21st-century computational geoscience research. This is a key part of CISL’s work to develop and maintain the highly skilled STEM workforce required for the continued advancement of science and technology.