New Supercomputing Fellowships

Postdoctoral students are the backbone of scientific institutions such as Los Alamos National Laboratory, as each new generation brings exciting ideas and fresh energy to the world of science. The Los Alamos Postdoctoral Program offers uniquely qualified early-career scientists the opportunity to perform challenging research in a scientifically rich research and development environment. Postdocs also have the opportunity to present and publish their research and advance knowledge in critical scientific areas at Los Alamos.

"I once served as part of the postdoc committee here at the Laboratory," explained Brian J. Albright, a scientist who works for the Plasma Theory and Applications Group at Los Alamos. "One of the things that was apparent to me was that this program is outstanding at bringing in fresh talent from the outside—the best scientists in the world in the areas of theoretical science or experimental science. But often, it seemed as if computational science was not served quite as effectively, so I wanted to develop a fellowship that specifically targeted computational and computer scientists to join the community here at the Laboratory. This community is very important, given that much of our weapons work today is done on computers."

In collaboration with other Los Alamos scientists and managers, as well as the Los Alamos Postdoc Office, Albright, along with Cheryl Wampler (Advanced Simulation and Computing Program) and Mary Anne With (Education and Postdoc Office), this year inaugurated the Metropolis Fellowship in Computer and Computational Science. Recipients of this fellowship can pursue advanced research in the areas of computational and computer science, physics, and engineering. Under the Advanced Simulation and Computing Program, computer simulation capabilities are developed to support the Stockpile Stewardship Program, as well as broader national security needs.

Selecting Candidates

The competition for this year's inaugural fellowship was rigorous. The Los Alamos Fellowship Prescreening Committee reviewed candidates from around the world, as this fellowship is open to citizens and noncitizens. The pre-screening committee selected and advanced 12 candidates, who underwent two additional reviews through another screening committee consisting of representatives from the Computational Physics; Theoretical Design; Computer, Computational & Statistical Sciences; and High-Performance Computing division offices, as well as from the Advanced Simulation and Computing Program Office.

"Prospective postdocs must have a mentor who sponsors them," says Albright. "So, one of the functions of the prescreening committee was to match possible postdocs with prospective mentors."

Four offers were extended to candidates, two of whom were already postdocs at the Laboratory. All four Metropolis fellows are U.S. citizens.

This Year's Fellows

Adam Manzanares

Adam Manzanares works under the mentorship of Meghan Wingate and John Bent, both in the High Performance Computing Division. Manzanares grew up in northern New Mexico and attended the New Mexico Institute of Mining and Technology while working on a BS in computer science. From 2002 to 2007, Manzanares worked as a student intern at Los Alamos—he had the opportunity to work on wireless network security and integration. In the spring of 2010, he received his PhD in computer science from Auburn University, where he concentrated on energy-efficient storage systems. His research interests include high-performance and parallel computing, storage systems, and computer science education.

As a Metropolis Fellow, Manzanares is working to determine how a parallel log-structured file system (PLFS) will fit into an exascale input/output (I/O) stack and what improvements PLFS will require to operate at this extreme scale of computing.

As computer platforms move toward the exascale era, the complexity of a system will drive downward the mean time to interrupt (MTTI). As the MTTI begins to diminish, the frequency of checkpointing increases for parallel applications. Such checkpointing requires that the parallel file systems used to store checkpoint data deliver high-write bandwidths across a variety of I/O workloads. Shared file writing is one particular workload that is in heavy use, but this workload also causes performance degradation on many parallel file systems. Los Alamos computer scientists developed PLFS to map these difficult I/O workloads into manageable workloads on a parallel file system.

Parallel Log-Structured File System image

The Parallel Log-Structured File System (PLFS) decouples a logically shared file, This improves write bandwidth significantly and allows scientists to map the workload into parallel I/O optimized layouts. PLFS helps shield application developers from increasingly complex I/O subsystem details.

Although PLFS solves the shared-file-writing dilemma, scientists quickly realized that its performance required the improvement of read performance before it could go into production at Los Alamos. Manzanares has developed several collective I/O optimizations that have increased PLFS's read performance so that it rivals direct access to the parallel file system. Such enhanced performance in PLFS has enabled Los Alamos computer scientists to manipulate I/O metadata workloads so that it is possible to achieve high performance from a parallel file system.

Christopher Ticknor

With Laboratory Fellow Lee Collins as his mentor, Chris Ticknor works at the Laboratory's Physics and Chemistry of Materials Group. Ticknor is studying quantum correlations computationally in two systems. The first involves warm dense matter, in which electrons are treated quantum mechanically and the ions are treated classically. The second involves a multielectronic molecule interaction with an attosecond pulse. Both projects require the use of supercomputers.

Scientists postulate that warm dense matter is at the core of some large planets. It is also possibly within the solid-to-plasma phase transition driven by laser pulses during inertial confinement fusion and in other systems that start as solids but are heated to become plasmas. The latter two examples are relevant to nuclear explosions.

Ticknor was a Director's Postdoctoral Fellow at Los Alamos National Laboratory and previously was a postdoctoral fellow at Swinburne University of Technology in Melbourne, Australia. Ticknor holds a BS from Bucknell University and a PhD from JILA/University of Colorado at Boulder. Both degrees are in physics.

Jupiter interior photo

Scientists are studying warm dense matter, in part, to better understand planetary interiors (such as Jupiter's, shown here) in terms of equation of state and other material properties. Such understanding will enable scientists to one day understand and perhaps even predict planetary formation and evolution.

David Collins

Co-sponsored out of the Theoretical Design and Theoretical divisions at Los Alamos, David Collins is working under the guidance of principal investigators James Cooley, Hui Li, and Shengtai Li. Collins' focus area is on determining the role of magnetic fields in star formation. His work will involve developing models and implementing them into large-scale multiphysics computer codes.

David Collins began his studies at the Eastman School of Music in Rochester, New York. He then transferred to the University of Cincinnati to study engineering and ultimately graduated with degrees in mathematics and physics. While in Cincinnati, Collins discovered a deep interest in programming and experiment modeling while working with particle physics experiments. Collins worked with Professor Michael L. Norman on numerical magnetohydrodynamic methods for adaptive mesh refinement codes, as well as with Professor Paolo Padoan on studies of star formation and supersonic turbulence. Collins graduated in 2009 but remained at the university for two years as a postdoc.

At Los Alamos, Collins has begun working on extending the numerical models of star formation he developed in graduate school. He intends to include a broader and more realistic description of physics in these models. The turbulent fragmentation model Collins is using possesses well-described statistical properties of isothermal supersonic turbulence used to predict the formation rate and mass distribution of stars in the galaxy. He will follow collapsing protostars beyond the isothermal regime to include several changes of state in the magnetized hydrogen. He will also model the formation of isothermal molecular clouds from their larger atomic progenitors. Collins will use a combination of two codes: Enzo (developed at the University of California, San Diego) and RAGE (developed at Los Alamos).

Jupiter interior photo

The Orion Nebula (shown here), is an archetypical example of star formation. New research suggests that cosmic magnetic fields, which can channel condensing interstellar gas, may play an important role in the birth of stars.

Brian Haines

Scheduled to begin his fellowship in the summer of 2011, Brian Haines will work for the Los Alamos Computational Physics Division under the mentorship of principal investigator Fernando Grinstein. Haines' work will address fundamental turbulence-physics-modeling issues of advanced programmatic interest, such as the mixing of materials in shock-driven turbulence.

 

In this issue...

Contact Us | Careers | Bradbury Science Museum | Emergencies | Inside LANL | Maps | Site Feedback | SSL Portal | Training

Operated by Los Alamos National Security, LLC for the U.S. Department of Energy's NNSA © Copyright 2016 LANS, LLC All rights reserved | Terms of Use | Privacy Policy