Database Reference
In-Depth Information
lifetime to an unacceptable level. Anomalous tokamak plasma transport is
thought to be associated with small-scale plasma turbulence. When the heat-
ing power to the core plasma is above a certain threshold, a thin plasma
layer forms, making the plasma almost free of turbulence; in addition, the
central plasma temperature and density rise under these conditions with the
added benefit that the fusion probability increases dramatically. However,
this layer also triggers large magnetohydrodynamic-type instabilities, which
simultaneously destroy the layer (lowering the fusion power in the core) and
dump the plasma energy to the material walls. It is currently not fully un-
derstood how this layer builds up or how the following crash occurs. The
success of next-generation burning plasma experiments is heavily dependent
upon solving this problem. Thus, understanding these physical processes is
an important area in fusion plasma research (see Cummings et al.
25
for more
details).
Due to the complexity of running the simulation (e.g., staging input data,
monitoring execution status, and managing result data) as well as to the rapid
changes and evolution of the simulation code itself, automating the execution
of the simulation via a workflow is crucial for both XGC1 developers and sci-
entists wanting to evaluate XGC1 results. Here, we can distinguish three layers
of activity: (1) At the highest level, physicists are interested in understand-
ing (and ultimately taming) nuclear fusion in tokamak-type reactors; (2) in
addition to performing the actual experiments, sophisticated, large-scale sim-
ulations on supercomputers are used to gain insights into the process (this is
a goal of the CPES project); and finally, (3) a simulation management work-
flow is used to deal with the challenging issues in running the simulations and
automating the necessary steps as much as possible.
13.3.1 Overview of the Simulation Monitoring Workflow
The XGC1 simulation outputs one-dimensional diagnostic variables in three
NetCDF files. It also outputs three-dimensional data written in a custom bi-
nary format for ecient I/O performance. The latter data subsequently is
converted to standard formats, such as HDF5, for archiving and analysis.
Simple plots are generated from the 1D diagnostic variables, while 2D vi-
sualizations (i.e., cross-section slices of the tokamak) are produced from the
converted 3D data. The NetCDF, HDF5, and all images produced during the
simulation are archived.
Figure 13.1 shows a graphical representation of a CPES simulation mon-
itoring workflow implemented using the Kepler scientific workflow system.
18
After initial preparation steps (e.g., checking if a simulation restart is re-
quested; logging in to all involved machines; creating directories at the pro-
cessing sites), two independent, concurrently executing pipelines are started
for monitoring the XGC1 simulation. The term
monitoring
is used to indicate
that for each output step, plots and images are generated in real time, and
that these can be visualized on a dashboard, enabling the scientist to observe
