Database Reference
In-Depth Information
of the energized particles with the target. Each of these different models is
studied with one or more different simulation codes. One grand challenge in
accelerator modeling is to simulate the entire beamline, from injection to im-
pact. This ambitious project is being approached by the development and use
of a number of different codes, each of which models a different portion of the
beamline.
∗
Accurate end-to-end modeling of accelerators is crucial to create
an optimized design prior to building the multi-billion-dollar instrument.
Unfortunately, there has historically been little coordination between these
code teams: The individual codes often output particle and field data in dif-
ferent, incompatible formats. Worse yet, some of these codes output multiple
terabytes of simulation data in ASCII format because it is easy to do so. Many
of these codes use serial I/O; for example, ASCII, HDF4. As a result, there
is data format incompatibility between different modeling stages, I/O is slow
when performed in serial, and it is wasteful of space in the case of ASCII
formats. Of more concern is the impediment to scientific progress that results
from data format incompatibility. It becomes extremely dicult to compare
results from different simulations that model the same part of the beamline
due to different data formats (e.g., different units of measure, different coor-
dinate systems, different data layouts, and so forth).
H5Part
†
was motivated by the desire of the accelerator modeling commu-
nity to address these problems. The vision is to have a community-centric
data format and API that would enable codes from all processing stages to:
(1) read/write particle and field data in a single format; (2) allow legacy
codes—simulations, visualization and analysis tools—to quickly take advan-
tage of modern parallel I/O capabilities on HPC platforms; (3) accelerate
software development and engineering in the area of I/O for accelerator mod-
eling codes; (4) facilitate code-to-code and code-to-experiment data compar-
ison and analysis; (5) enable the accelerator modeling community to more
easily share data, simulation, and analysis code; and (6) facilitate migrating
toward community standards for data formats and analysis/visualization. The
intended result is to accelerate scientific discovery by reducing software de-
velopment time, by fostering best practices in data management within the
particle accelerator modeling community, and to improve the I/O eciency
of simulation and analysis tools.
H5Part is a very simple data storage schema and provides an API that
simplifies the reading/writing of the data to the HDF5 file format.
10
An im-
portant foundation for a stable visualization and data analysis environment is
a stable and portable file storage format and its associated APIs. The presence
of a common file storage format, including associated APIs, fosters a funda-
mental level of interoperability across the project's software infrastructure. It
∗
See http://compass.fnal.gov for an example of a large, community-based accelerator modeling
project with exactly these objectives.
†
We coined the term H5Part for the API to reflect use of HDF5 for particle-based datasets
common in high-energy physics.
