October 2019 | ChEESE - Centre of Excellence for Exascale Supercomputing in the area of the Solid Earth

News

ChEESE code developers join CINECA GPU Hackathon

22 October 2019

ChEESE partners recently participated in the CINECA GPU Hackathon which took place in CINECA Rome from October 7-10, 2019.

Lead developers Ravil Dorozhinskii and Lukas Krenz from the SeisSol team and Natalia Gutierrez and Leonardo A. Mingari from the FALL3D team were there to work on porting and optimizing their codes. Highly qualified NVIDIA mentors worked with the developers side by side in order to help their codes reach Exascale capability.

"I believe that all this information and experience gained during the hackathon are going to help us in development of SeisSol. The new profiling data have already changed our general view and the current strategy to make our code more efficient," said Dorozhinskii.

News

How to build a society resilient to natural catastrophes: ChEESE team defines HPC service prototypes

20 September 2019

Image: Simulated maximum surface elevation due to an earthquake generated tsunami offshore Japan. The white points indicate tsunami DART stations that monitor the tsunami propagation. In ChEESE, we will simulate such events faster than the tsunami propagates over the ocean, meaning that the simulations are carried out faster than the speed the tsunami propagate.

Imagine these scenarios in which natural disasters could potentially threaten hundreds or even thousands of lives. What research is currently being done and what role does ChEESE play in this?

A tsunami wave generated by a strong earthquake, landslide or volcanic eruption could propagate across the Mediterranean Sea in only few hours. Can we provide faster than real time inundation maps at high resolution, and a quick alert also for the populations at risk?
What would happen if a large earthquake occurred in Europe? Can scientists provide ground motion and shaking maps fast enough and at sufficiently high resolution to help emergency management and rescue operations?
When the next explosive eruption in Iceland occurs, will scientists be able to provide reliable, real-time ash distribution in the atmosphere to aviation authorities, to avoid massive closure of the airspace while guaranteeing the safety of air routes?

In cases of a natural catastrophes, efficient computational algorithms and the most powerful computer in Europe might be required to help manage these emergencies, in the shortest possible time.

This is Urgent Computing. ChEESE aims to demonstrate in Pilot Demonstrators (PD) on Urgent Seismic Simulations (PD1), Faster Than Real-Time Tsunami Simulations (PD2) and High-Resolution Volcanic Ash Dispersal Forecast (PD3) that technology is ready to design Exascale Urgent Computing workflows for supporting contingency plans for seismic, volcanic and tsunami events.

In addition, natural catastrophes often occur with very little anticipation and precursors difficult to detect. The earlier the warning, the more effective the mitigation of the impact of the dangerous phenomena.

This is Early Warning. ChEESE will make use of the most efficient High Performance Computing architectures, software and workflows to demonstrate prototype Early Warning systems for tsunamis .

How big will the next volcanic eruption be at Jan Mayen, in Norway? How will that potentially impact the population and the air traffic in Europe, considering the statistical variability of wind intensity and directions?
Can scientific research, together with advanced computing technology, help reduce volcanic risk for the population and for productive activities in Southern Italy and optimize the land use, in consideration of the potential impact of future volcanic eruptions at Campi Flegrei and Vesuvio?
What is the probability of ground acceleration exceeding a given threshold at a nuclear plant site, or for a critical infrastructure in Europe? What is the probability that a large earthquake in the Mediterranean produces a tsunami wave higher than one metre in the Marseille harbour?

Building a prepared society resilient to natural catastrophes requires the capability of managing the complexity of the natural phenomena and the large uncertainty associated with their development, in a probabilistic framework. This requires performing large ensembles of accurate scenario simulations to reproduce the complex physics of the natural systems and the wide variability of initial and boundary conditions.

This is Probabilistic Hazard Assessment. ChEESE will demonstrate, with Pilot Demonstrators on Physics-Based Probabilistic Seismic Hazard Assessment (PD59, Probabilistic Volcanic Hazard Assessment (PD6) and Probabilistic Tsunami Hazard Assessment (PD7), the capability of Exascale Computing to perform large ensemble, physics-based simulations to quantitatively assess natural hazards and their related uncertainties.

(*) Pilot Demonstrators (https://cheese-coe.eu/results/pilots) are prototype applications in the field of seismology, volcanology and tsunami research that make use of massively parallel architectures available to ChEESE at European Supercomputing Centres to provide scalable workflows, for the benefit of the scientific community and of the whole society.

Scientists yearn to know,

society needs to know.

News

SPECFEM3D can now simulate seismic waves twice as fast

11 October 2019

Image 1: Simulation of seismic waves propagation. (From: https://geodynamics.org/cig/software/specfem3d/)

SPECFEM3D is well-known code in the seismology community. It is a code that simulates 3D seismic wave propagation in any region of the Earth based on the spectral-element method. In fact, it was one of the first seismological simulation codes capable of processing 3D models on parallel computers. It can handle a multitude of topographies and numerical models (elastic, viscoelastic, acoustic, poroelastic, anisotropic, etc.). It was initially used to simulate the peak ground motion taking into account whole 3D geological complexity such as topography, basins and internal discontinuities. This allows the production of precise shaking intensity maps that play an important role in seismic hazard management. SPECFEM3D is also used in inverse problem workflow for subsurface imaging at different scales from global earth to seismic exploration. It is increasingly used in others areas such as volcanology and also at smaller scales in non-destructive testing and ultrasound medical imaging.

As part of the CHEESE project, SPECFEM3D has been selected for two Pilot Demonstrators: Pilot Demonstrator 1: “Urgent Seismic Simulations” and Pilot Demonstrator 9: “Seismic tomography”.

As a partner of the project, BULL/ATOS is working on SPECFEM3D’s software optimization. Through the optimization work of BULL/ATOS, SPECFEM3D can now take advantage of modern x86 SIMD (Single Instruction Multiple data) instructions sets such as AVX2 or AVX512. These optimizations allow higher processing rates by allowing each core to handle multiple computations simultaneously. This has already allowed the code to obtain a speedup of 2x on one of CHEESE test cases on ATOS’ BullSequana machine equipped with Intel processor Skylake 6148, thus enabling a faster execution of the Pilot Demonstrators.

The speed-up achieved will allow us to provide near-real-time maps of ground motion and shaking intensity following significant earthquakes. These maps could be used by different organizations, for post-earthquake response and recovery, public and scientific information.

For the seismic tomography Pilot Demonstrator the speed-up of the simulation part of the workflow is a major advance. It will allow to use more data at higher frequencies in order and then produce higher resolution subsurface images.

SPECFEM3D was developed in the early 2000s through a collaboration between Jeroen Tromp from Princeton University and Dimitri Komatitsch from the French National Center for Scientific Research (CNRS). Sadlly, Dimitri passed away in January 2019. He will be deeply missed by all those whose lives he touched, and his legacy will live on in this project that has inspired us all.