Data-streaming workflow for seismic source location with PyCOMPSs parallel computational framework

Modern digital seismic networks record a wealth of high-quality continuous waveforms that contain a variety of signals associated to a wide range of seismic sources (e.g., earthquakes, volcanic, tectonic tremors, environmental sources) that probe transient energy release processes. Efficient and automatic detection, location and characterization of these different seismic sources is critical to understand slowly-driven evolution of active tectonic and volcanic systems toward catastrophic events. Developing a common analysis framework for systematic exploration of the increasing wealth of seismic observation streams is important for improving seismic monitoring systems and extracting large and accurately resolved seismic source catalogues.

To this end, we present a scalable parallelization with PyCOMPSs (Tejedor et al., 2017) of the python-based BackTrackBB data-streaming workflow (Poiata et al., 2016; 2018) for automatic detection and location of seismic sources from continuous waveform streams recorded by large seismic networks. This allows achieving an efficient distribution and orchestration of BackTrackBB code on different architectures. PyCOMPSs is a task-based programming model for python applications that relies in a powerful runtime able to extract dynamically the parallelism among tasks and executing them in distributed environments (e.g. HPC Clusters, Cloud infrastructures, etc.) transparently to the users.

We will provide details of the PyCOMPSs-based BackTrackBB workflow implementation. Results of scalability tests and memory usage analysis will be also discussed. Tests have been performed, in the context of the European Centre Of Excellence (CoE) ChEESE for Exascale computing in solid earth sciences, on the MareNostrum4 High-Performance computer of the Barcelona Supercomputing Centre, using large-scale datasets of synthetic and real-case seismological continuous waveform data sets.