CISL offers Raspberry Pi parallel computing projects for students

By Staff
08/15/2014 - 12:00am

On 19 May 2014 CISL welcomed its first cohort of students to a new project designed to teach parallel computing principles using inexpensive Raspberry Pi (R-Pi) computers. Administered under CISL’s SIParCS (Summer Internships in Parallel Computational Science) program and funded by NCAR’s Diversity Committee, this pilot program introduces supercomputing concepts to undergraduate students at U.S. Minority Serving Institutions (MSI).  Both Hampton University and Salish Kootenai College, this year’s home institutions, were selected after faculty showed interest in the pilot as recipients of CISL-funded travel scholarship for MSI faculty at the 2013 Front Range Consortium for Research Computing (FRCRC) High-Performance

Lauren Patterson and Mentors

Lauren Patterson (right, Hampton University) and her teacher assistant, Amogh Simha (left, University of Colorado, Boulder) discuss cluster computing with project lead Rich Loft (standing).

Computing Symposium, now Rocky Mountain Advanced Computing Consortium’s (RMACC). The R-Pi platform is a small, very low-cost computer that runs the full Linux operating system. The R-Pi’s “system on a chip” processor uses the same standard architecture as 95% of smart phones, 35% of digital televisions, and 10% of mobile computing devices. The R-Pi platform removes the cost barriers for smaller colleges and universities to teach and conduct research in the principles of high performance computing (HPC). Students used the R-Pi to construct miniature clusters, create software environments analogous to those found at major supercomputing centers, and deploy applications (including simplified geoscience applications) for only hundreds of dollars.

A unique extension to CISL’s successful SIParCS program, this new program supports talented nontraditional students who would otherwise be unable to participate in an 11-week internship program onsite at NCAR. Using the R-Pi platform provides diversity and education-related benefits in several ways. First, students participating in the summer 2014 pilot gained access to new technology, skills, and insight into possible educational and career paths. This new program recognizes the various responsibilities and commitments of nontraditional students while providing the resources of a traditional internship. In addition, since this program allows interns to spend seven of their 11 weeks at their home institution, it addresses barriers such as work or family responsibilities, lack of adequate information about postsecondary opportunities, or negative perceptions of academic ability.

CISL recently hired Raghu Raj Prasanna Kumar and Amogh Simha as teacher assistants (TAs) for the R-Pi project. Both are graduate students working on the R-Pi pilot project with Dr. Richard Loft (Director of CISL’s Technology and Development Division), Stephanie Barr (CISL Diversity Coordinator), and Karina Hauser (a visiting professor from Utah State University). Both TAs provided our first R-Pi students with the technical support needed to complete their summer projects. Each project centered on learning modern supercomputing paradigms. In their first weeks, they helped assemble four-node R-Pi clusters, performed system configuration, and installed software for the Raspian Linux distribution, MPI, and various tools (benchmmark tests) for analyzing system performance.

Justin Moore and Mentors

Extern, Justin Moore (middle), updates CISL Diversity Coordinator, Stephanie Barr (right) on his daily tasks with teacher assistant, Raghu Raj Prasanna Kumar (left seated).

The students returned to their home institutions to continue work on their projects in the second week of June. Their CISL mentors and TAs helped them and their local mentors complete two unique projects:

• Exploring and analyzing simple, advanced, and complex datasets of varying sizes. These real-world datasets are taken from the environmental sciences. This project used Hadoop MapReduce-based databases. MapReduce is a programming model for processing and generating large data sets with a parallel, distributed algorithm on a cluster computer. Hadoop MapReduce is a software framework for easily writing applications that process vast amounts of data in parallel on large cluster computers in a reliable, fault-tolerant manner.

• Studying the R-Pi cluster computer’s performance using a variety of standard HPC benchmark tests: Linpack, message passing performance tests, and I/O benchmarks. This traditional HPC work is designed to increase student understanding of parallel programming and program optimization.

In July 2014, CISL mentors visited the students’ home institutions to assist with projects. The students recently returned to NCAR during the final week of the SIParCS program to give oral presentations about their work during the SIParCS colloquium. We are pleased to report that both students had posters accepted for the 4th annual RMACC High-Performance Computing Symposium. Each presented their findings Aug. 12-13 on the University of Colorado Boulder campus.