-- EricBohm - 26 Feb 2008

The goal of our new grant Scalable Atomistic Modeling Tools with Chemical Reactivity for Life Sciences is to meld together OpenAtom (AKA leanCP) with NAMD so that simulations of biological systems can be extended to include the study of systems where accurate simulation of the electronic structure is necessary for part of the system, while the bulk of the system can be studied using classical mechanics.

The principle investigators are LV Kale, Mark Tuckerman, Jack Dongarra, and Glenn Martyna. Work will occur primarily under the purview of: Glenn Martyna @ IBM, Mark Tuckerman @ NYU, Eric Bohm @ UIUC, and Jakub Kurzak @ UTK. Email All

Doing this efficiently requires a multi-pronged effort:

• QM/MM algorithms integrating NAMD with OpenAtom
  • Dual Grid Euler Exponential Spline
• Sequential algorithms
  • 3D FFT
  • non-square DGEMM
• Parallel efficiency
  • Fast collectives
  • Phased hierarchical load balancing
  • Communication optimizations specific to the target architecture (Cray XT3/4/5)
• New algorithms
  • DVR Real space methods
• New OpenAtom features
  • Path Integrals
  • Parallel tempering
• New NAMD features
  • modular interaction for QM region
  • specific treatments for water
  • reduced memory footprint

Workplan for the next six months QMMMWorkPlan2008

ORNL System

ORNL Jaguar Cray XT AMD Dual Core Cores: 11,706 Peak: 119.4 TF Upgrading 250 TF Memory: 46 T

Phoenix Cray X1 Cray Vector Cores: 1,024 Peak: 18.3 TF Memory: 2 TB

ORNL Computer System Road Map http://nccs.gov/computing-resources/jaguar/roadmap/

OpenAtom performance on Cray

Performance of OPENATOM on BigBen? XT3. The numbers represent time per step in seconds. As of Feb 2008.

Recent Publications

Our IBM Journal Paper has our published scaling results for BG/L, along with a description of the parallel implementation. The parallel decomposition/flow of control diagram presented therein is linked below.

Shawn Brown recommends that QMMM be pronounced "queue mmmmmmmm"

-- EricBohm - 03 May 2008