a national resource for computational science education

HOME BWPEP Shodor Blue Waters

View Position
Project TitleParallel Program Analysis Intern Aiming for Petascale Performance (PPAIAPP)
SummaryThe first purpose of this internship is to instrument performance profiling tools into a newly parallelized electronic structure calculation program called the orthogonalized linear combination of atomic orbitals (OLCAO) method. Then, the goal will be to analyze the profiling data produced by the tools to give insight regarding performance bottlenecks and scaling on a variety of parallel computing platforms such as small to medium sized clusters as well as XSEDE-based high performance computers. This work will be performed in the context of a larger goal that aims to make the program take advantage of other routes to parallelization (e.g. co-processor based parallelization on the Intel Xeon Phi).
Job DescriptionIn this position, the intern will primarily be working on an ab initio density functional theory based program for computing the electronic structure of solids that has been developed at UMKC. The program is called the Orthogonalized Linear Combination of Atomic Orbitals (OLCAO) method. This program was recently parallelized within a message passing paradigm using MPI. The main purpose of the internship is to use performance monitoring and analysis tools to more deeply understand how the now parallelized application performs and scales for a wide variety of problem classes and on a variety of computer architectures. The intern will collect this data, organize it, and analyze it to find places where the performance of the code is lacking and to help guide further efforts directed at thread-level (OpenMP), general purpose graphics processing unit (GPGPU) based, and co-processor based parallelization.
The program is written in Fortran 90 and we will use various performance monitoring tools such as Tau, Papi, and IPM. The intern will be directed by the principal investigator and will be paired with an experience graduate student that has extensive hands on knowledge of the program code. The program code will initially be run on regional computers run by the Missouri research network (both cluster type and SMP), but as the project progresses there will be a shift to make use of Maverick and Stampede XSEDE machines for GPGPU and MIC architecture based parallelization.
This project is designed particularly for a student who is at or near an entry level of parallel programming knowledge and skill. The student will gain valuable experience (programming, debugging, parallel algorithm planning, etc.) on a challenging, but narrowly defined project so that there is little likelihood of being overwhelmed by technical complexity. Further, because the first complete MPI-based parallelization of the OLCAO code has just recently been achieved this will be an excellent project to initiate the long term involvement of the intern into further parallel program development (OpenMP, GPGPU, Xeon Phi) and the scientific application of the program. Relatively little experience is required so there is plenty of opportunity for on-the-job training. Self-motivation is critical.
Conditions/QualificationsQUALIFICATIONS: The most important qualification for this position is that the student needs to have a high degree of self-motivation and enthusiasm for the project. A good dose of patience and persistence will also be helpful. A background that includes experience with computers, basic computer programming, and getting around on a command line interface will be very helpful for making the most efficient use of the internship time. However, those skills are not essential prerequisites as long as the student has good analytical problem solving skills and a capacity to find, read, digest, and make use of technical literature. The student will also be working in a team environment and so good scheduling and coordination skills will be emphasized.

CONDITIONS: The student should commit to self-study of the relevant technical and scientific topics to gain basic familiarity before the internship start date. The student must also be enrolled as an undergraduate in the Physics and Astronomy BS program at UMKC.
Start Date05/25/2014
End Date05/31/2015
LocationCity/State: Kansas City, Missouri
Research Group: Computational Physics Group
Department: Physics and Astronomy
Institution: University of Missouri - Kansas City (UMKC)
William Leverette