New codes with ParaDyn: parallel-computing version of DYNA3D

PARADYN, a breakthrough computer code for parallel machines, was ready for action--the real action that only the most powerful of supercomputers could provide. Action began late last year when the IBM SP2 arrived, the first of a series of massively parallel supercomputers for Lawrence Livermore. Within days, ParaDyn was loaded on the machine and used to simulate, for the first time, a one-million-element model of a ground shock--a very large problem. The first simulation immediately demonstrated the machine's great power and potential, as well as the efficacy of ParaDyn.
New codes like ParaDyn are an important component of Livermore's effort to ramp up to 100-teraflops (100 trillion floating-point operations per second) power, requisite to solving problems for DOE's Accelerated Strategic Computing Initiative. ParaDyn is the parallel-computing version of DYNA3D, one of Livermore's premiere codes for modeling and predicting thermomechanical behavior.
The development of ParaDyn, ongoing for the past six years, was performed not just in anticipation of parallel computing needs, but because of the forward momentum provided by computational science in the Methods Development Group of Lawrence Livermore's Engineering Directorate. The group was formed in the 1970s to develop modeling tools that were critically needed by Laboratory nuclear weapons projects but were commercially unavailable (see box below). The broad capabilities of these tools--for analyzing the complex system deformations and nonlinear material interactions--made them valuable commercial products. Over the next 20 years, the group pursued cutting-edge code development and expanded the modeling technology base. They produced an entire family of state-of-the-art software for nonlinear analysis of thermomechanical systems.
And the work continues. As long as the Laboratory's scientific knowledge is needed for weapon systems and other extremely large systems advancements, the group must continue to develop and improve codes. The tools they produce must satisfy the Laboratory's analysis and computational performance requirements. Peter Raboin, leader of the group, puts it this way: "If we end up merely duplicating the work of others and, worse yet, they are solving the same problems faster and better, then we might as well close up the shop." So the group forges ahead, adding to the code family and making existing codes more robust, more accurate, faster, and capable of ever more applications.

cont here:

https://www.llnl.gov/str/Raboin.html