Combining HPC with Data-Intensive Research via UNICORE for Seismological Applications

Today advances in computational seismology require to realize complex geophysical simula-
tions in order to predict consequences of seismic phenomena by providing accurate mathemat-
ical models of wave and rupture propagation in complex 3D media. Data collected by seismic
stations must be analyzed and compared with results of simulations in order to gain new infor-
mation about Earth’s composition and subsoil in a fast and efficient manner. In order to provide
high-level tools to run complex simulations as HPC jobs and, at the same time, processing both
synthetic data and sensor data, geoscientists need to be supported by a wide distributed scien-
tific infrastructure able to ensure transparent access to resources. Within the VERCE
1
project
this goal is addressed exemplifying such challenges and implementing specific use cases, pro-
viding end users with software tools and application environments and defining a basic service-
oriented architecture. UNICORE comes in handy providing a software layer capable to run
complex workflows and supporting end users with graphical tools and GridBeans to modelize
complex simulation jobs and run them on the HPC resources. In this paper we describe how
UNICORE can be used in VERCE to enhance scalability of experiments combining the HPC
file-based model with data-processing operations. This is achieved by equipping the CINECA
UNICORE PLX installation with a Python extensible software layer. This consists in a suite
of Python scripts relying on the ObsPy Python library. These scripts perform the processing of
large amounts of data coming from different information sources, as for example seismic sta-
tions or HPC sites, managing data as part of a continuous job execution/data processing loop.