Adaptive
Computations for Fluids in Biological Systems
Project Leader:Kathy Yelick, UC Berkeley
Project Manager:Kathy Yelick, UC Berkeley
 |
Heart Simulations
Using the Immersed Boundary Method
This
figure shows the flow pattern in a thin section through the
model heart. The flow pattern is represented by streamlines
that are calculated from the 3-D velocity field at the instant
shown. Flow passes from the left atrium, through the mitral
valve (top center of the figure), and into the left ventricle.
The aortic valve (upper left) is closed at this moment.
|
The purpose of the Adaptive Computations alpha project is to develop
easy-to-use software tools to simulate fluid flow in biological
systems such as the human heart. Heart simulations play an important
role in understanding both healthy and diseased hearts. Simulations
can also be used in the development of prosthetic devices, such
as artificial heart valves, less invasive surgeries, and education.
Considerable computation on high-performance systems is needed
to conduct this research. For example, one simulation of a single
heartbeat required 100 CPU-hours on a Cray T90. That simulation
used the immersed boundary method, which models a biological system
as a set of elastic fibers within an incompressible fluid. This
simulation method is best known for its use to simulate fluid flow
in the human heart by developers Charles Peskin and David McQueen,
mathematics professors at New York University (NYU)and co-winners
of the 1994 Cray Research Leadership Award for Breakthrough Computational
Science. The immersed boundary method has been used to simulate
platelet coagulation during clotting, embryo growth, insect flight,
and other biological systems.
To port the code to the IBM Blue Horizon at the San Diego Supercomputer
Center (SDSC), participants use the Titanium language and compiler,
communication and cache optimizations developed by the Titanium
group, and new scalable solver technology from New York University.
A parallel implementation will enable the method to be used for
much more demanding problems.
The past year saw the first implementation of the immersed boundary
method that runs on distributed-memory, parallel platforms using
Titanium. The Titanium version of the software (TIBM) is based on
a generic version of the method, which separates application-specific
features into modules, making it easier to maintain and extend them.
Participants also instantiated the generic code with the features
necessary to simulate the heart, as well as software to partition
the heart model input across processors and to change the file format
to make it consistent with the generic software and improve the
I/O performance. In addition, Peskins group developed an OpenGL
version of the heart visualization software to replace the previous
implementation that relied on SGIs proprietary graphics library.
The full heart model has been run on several parallel machines,
including Blue Horizon, a Cray T3E at the National Energy Research
Scientific Computing Center (NERSC), an SGI Origin at the National
Center for Supercomputing Applications, and the Millennium cluster
at UC Berkeley. These runs have demonstrated the portability of
Titanium.
The primary goal for the coming year is to improve the performance
of the TIBM code, support a second application, and add features
necessary for applications other than the heart. The target application
is to model a collapsible tube, which exhibits patterns of oscillation
under certain conditions, but for which the physics is not understood.
This is needed to understand how arteries behave when a blood pressure
cuff is applied and how accurately that measurement corresponds
to actual pressure in the artery. Researchers at NYU are developing
the artery model, and the UC Berkeley group will help add the model
to the TIBM code. The alpha project team also will explore the use
of the Active Data Repository to support the analysis of 3-D data
sets from cardiac blood flow studies.
|