Presentation
Visualizing the Atomic Detail Dynamics of Biomolecular Complexes in Our Compute-Rich but I/O-Constrained Future
SessionVisualization and HPC
Speaker
Event Type
Focus Session
Big Data Analytics
Heterogeneous Systems
Molecular Research
Scientific Software Development
Visualization & Virtual Reality
TimeMonday, June 25th5:20pm - 5:40pm
LocationPanorama 3
DescriptionThe combination of experimental structure information,
molecular dynamics (MD) simulation, and high-fidelity visualization
techniques with HPC computing resources creates a powerful
"computational microscope" that permits molecular scientists to view the
structure and dynamics of biomolecular complexes in atomic detail
within realistic cellular environments.
The increasing size and time scale of these simulations presents many
challenges for simulation preparation, analysis, and visualization.
Despite the tremendous growth in the arithmetic performance of HPC systems,
generally more modest gains have been made in the performance and capacity
of memory and storage systems. HPC technologies such as remote visualization,
in-situ and in-transit analytics and visualization, burst buffers and NVDIMMs,
NVLink-interconnected GPUs, and so-called "fat node" designs are each
important tools for addressing the data-centric challenges we currently face.
The remaining challenges must be met with new algorithms, and in some
cases with unconventional thinking.
I will describe some of the key analytical and visualization challenges
that have arisen for state-of-the-art MD simulations and methodologies,
how the strengths of next-generation HPC systems can be used to overcome
their weaknesses, and the many adaptations of biomolecular visualization
tools such as VMD that are required to address these challenges successfully.
molecular dynamics (MD) simulation, and high-fidelity visualization
techniques with HPC computing resources creates a powerful
"computational microscope" that permits molecular scientists to view the
structure and dynamics of biomolecular complexes in atomic detail
within realistic cellular environments.
The increasing size and time scale of these simulations presents many
challenges for simulation preparation, analysis, and visualization.
Despite the tremendous growth in the arithmetic performance of HPC systems,
generally more modest gains have been made in the performance and capacity
of memory and storage systems. HPC technologies such as remote visualization,
in-situ and in-transit analytics and visualization, burst buffers and NVDIMMs,
NVLink-interconnected GPUs, and so-called "fat node" designs are each
important tools for addressing the data-centric challenges we currently face.
The remaining challenges must be met with new algorithms, and in some
cases with unconventional thinking.
I will describe some of the key analytical and visualization challenges
that have arisen for state-of-the-art MD simulations and methodologies,
how the strengths of next-generation HPC systems can be used to overcome
their weaknesses, and the many adaptations of biomolecular visualization
tools such as VMD that are required to address these challenges successfully.
Speaker
