UNIVERSITY OF TENNESSEE
Grant: $1,330,000 - National Science Foundation - Jun. 15, 2009
No votes have been cast for this award yet
Join the conversation: Post a comment about this award
Award Description: We propose to develop the core of a new generation of scalable chemistry and physics codes with an immediate target of the petascale computers now being deployed by the NSF but with a vision that incorporates the exascale computers anticipated within a decade. The intellectual merit of our proposed work we intend to be in the areas of general computational science and in chemistry and physics in particular. Extensive experience with codes designed for the previous generation of terascale computers has demonstrated their general inability to scale up to hundreds of thousands of processors. This failure is usually not inherent to the physics but due to the expression of limited amounts of parallelism and the use of data structures, algorithms and programming models that are tied to a specific granularity of parallelism (e.g., a certain loop nest) and to a specific machine model. In addition to these problems, mainstream parallel programming models do little to relieve the scientist or engineer from the burden of placing data and scheduling computation in an increasingly complex computer architecture. This escalating complexity is inhibiting both discovery in fields for which simulation is an established tool and application of high-performance computation by new users or to new topics. Starting from the MADNESS scientific simulation environment we will introduce new abstractions and increase the virtualization of the programming model away from the underlying hardware into a non-process centric model. An intelligent runtime environment with detailed scheduling and performance models will relieve the programmer of much of the burden described above, enabling focus upon the key elements of correct science and expression of all available concurrency. A domain-specific framework for molecular electronic structure and many-body solid-state physics and will be developed upon these tools motivated by the success of NWChem and similar frameworks in enabling science in the terascale era. We anticipate that the broader impact of our project will be in both enabling new science and in opening up petascale computers to a much wider audience. In addition to providing new simulation capabilities in molecular and solid-state electronic structure at the petascale, the computational environment will be broadly applicable and inter-operable with existing programming models to extend the lifetime of legacy code.
Project Description: Internal financial systems have had accounts created for each department to charge to, and work is now underway. We have commenced project planning and coordinating activities between the three groups (Harrison, Eguiluz, Peterson), the corresponding departments (chemistry, physics, electrical engineering and computer science) and the Joint Institute for Computational Science. A set of working equations has been produced for the multi-configuration self-consistent field code and work has commenced on the initial implementation within MADNESS with these updates committed to the public, open-source code repository at http://code.google.com/p/m-a-d-n-e-s-s/. We are searching for a new postdoctoral research assistant in the physics department, and plan for the position to be filled in the coming quarter. So far we have interviewed one candidate. We are also considering a candidate for a postgraduate research assistant position in the Joint Institute for Computational Science, and plan to complete the hiring in the coming quarter.
Infrastructure Description: This is a conditional field. This grant is not an Infrastructure project.
Jobs Summary: One new graduate students in the department of electrical and computer engineering is now being supported by this project. (Total jobs reported: 0)
Project Status: Less Than 50% Completed
This award's data was last updated on Jun. 15, 2009. Help expand these official descriptions using the wiki below.
No comments have been added for this project.