Julia for HPC

Julia for HPC

Many researchers may face this problem developing numerical applications. We prototype using a high-level language appealing for its eye to program, readability, nice plotting, very talkative debugger. When it comes to productions runs, we like to translate the prototypes in lower-level compiler languages to benefit from runtime performance, parallelisation possibilities but loosing many interesting features from high-level languages.

Our contribution presented at JuliaCon 2019 (Baltimore MA, USA) is an illustration of Julia solving “the two language problem”. We replace our Matlab prototype and the CUDA C + MPI production code by a single Julia code that serves both prototyping and production tasks. We showcase the port to Julia of a massively parallel Multi-GPU hydro-mechanical stencil-based solver in 3-D. The iterative solver can be applied to a wide range of coupled differential equations.

Figure 1. Weak scaling of the parallel GPU MPI hydro-mechanical solver. We report the parallel efficiency for both the Julia and the CUDA C implementation on 1024 and 5200 GPUs (full machine) respectively, on the Piz Daint hybrid Cray XC 50 at the Swiss National Computing Centre, CSCS.

We report a close to optimal weak scaling on 1024 NVIDIA Tesla P100 GPUs on the hybrid Cray XC-50 “Piz Daint” supercomputer at the Swiss National Supercomputing Centre, CSCS (Figure 1). We compare these results obtained with our Julia prototype to a reference scaling realised using the Multi-GPU production code solver written in CUDA C + MPI that achieved a high performance and a nearly ideal parallel efficiency on up to 5120 NVIDIA Tesla P100 GPUs “Piz Daint”. Soon in press.

Nvidia GTC 2019

Nvidia GTC 2019

This year’s Nvidia GPU Technology Conference to take place in San Jose, Silicon Valley, CA. Besides the opening keynote by CEO Jensen Huang, a former researcher from the Swiss Geocomputing Centre, Ludovic Räss, gave a talk on geo-supercomputing. Recording is accessible hereafter or on GTC on-demand:

RESOLVING SPONTANEOUS NONLINEAR MULTI-PHYSICS FLOW LOCALISATION IN 3-D: TACKLING HARDWARE LIMIT

For further information, feel free to contact lraess[at]stanford.edu.

Resolving thermomechanical coupling in two and three dimensions

Resolving thermomechanical coupling in two and three dimensions

Thibault Duretz, Ludovic Räss, Yury Podladchikov and Stefan Schmalholz recently published a new study about multi-physics couplings with focus on thermomechanical interactions.

Thermomechanical strain localisation in 3-D for a cylindrical and spherical inclusion. The solution is obtained using a pseudo-transient GPU-based solver.

Their contribution (accessible here) assesses the ability of an iterative technique to resolve nonlinear interactions between thermal and mechanical processes. This new approach to solution is particularly well suited for parallel devices, such GPUs. The authors benchmark their proposed solver against a direct-iterative type of more classical approach (TM2Di).

The codes illustrating and supporting this work can be accessed on the software page, or on the authors Bitbucket repository.

Conference Jose Fernando Mendes, 14 Nov 2018

Conference Jose Fernando Mendes, 14 Nov 2018

Jose Fernando Mendes
Professor at the University of Aveiro

Structural Properties of Multiplex Networks

Many complex systems, both natural, and man-made, can be represented as multiplex or interdependent networks. Multiple dependencies make a system more fragile: damage to one element can lead to avalanches of failures throughout the system.
In this talk I will present recent developments about the structural properties of multiplex networks. The transition founded is asymmetric. It is hybrid in nature, having a discontinuity like a first-order transition, but exhibiting critical behavior, only above the transition, like a second-order transition. A complete understanding of the transition cannot therefore be had without first understanding this critical behavior. I will discuss and describe the nature of such hybrid phase transitions and the appearance of avalanches at criticality.
José Fernando F. Mendes is a theoretical physicist working on Statistical Physics. His research focuses mainly in the study of complex systems and the structure and the evolution of complex networks like the World Wide Web, the Internet, biological networks, etc. Other interests are related with: granular media, self-organized criticality, non-equilibrium phase transitions, deposition models,etc.
He is co-author of over 130 scientific papers receiving about 18,000 citations, with his most cited works more than 3,000 citations.

Ludovic Räss | Spontaneous formation of fluid escape pipes resolved in 3D

Ludovic Räss | Spontaneous formation of fluid escape pipes resolved in 3D
“Ubiquitous observations of channelised fluid flow in the form of pipes or chimney-like features in sedimentary sequences provide strong evidence for significant transient permeability-generation in the subsurface. Understanding the mechanisms and dynamics for spontaneous flow localisation into fluid conductive chimneys is vital for natural fluid migration and anthropogenic fluid and gas operations, and in waste sequestration.”

 

High-permeability chimney genesis out of a source region in three dimensions. Colour plot (logarithmic scale) of dynamic permeability for two different lithologies, conductive sandstone and impermeable shale. Contoured values show a 1.5 order of magnitude increase in representative for the chimneys.

 

These motivations resulted in the development of a high-performance computing (HPC) application to resolve in 3-dimensions a coupled hydro-mechanical model. Very high spatial and temporal resolution is required in order to capture with accuracy the significant flow localisation in space and time. Ludovic Räss, Nina S.C. Simon and Yury Y. Podladchikov relied on the computing power delivered by the GPU-based supercomputer octopus, hosted by the Swiss Geocomputing Centre at the FGSE (Unil) and performed parallel simulations on 128 GPUs, i.e. more that 380’000 cores. This novel results permit to shed light on a previously unidentified fluid focusing mechanism that has a profound impact on assessing the evolution of leakage pathways in natural gas emissions, for reliable risk assessment for long-term subsurface waste storage, or CO2 sequestration.

Link to publication


Some additional movies are presented here:

Chimney formation mechanism. Three successive time laps of two-dimensional vertical (a–d) and horizontal (e,f) slices. (a,e) Dynamic permeability (logarithmic scale) field. The white arrows represent the fluid flux vectors, scaled by the maximal flux over time and directed into the chimney in the local drainage area, showing flux from outside to inside the chimneys. (b,f) Strain rate-dependent non-linear bulk viscosity values (logarithmic scale). (c,g) Effective pressure distribution. (d,h) Shear stress deformation magnitude (second invariant of the deviatoric stress tensor). White contour lines (b–d,f–h) represent the chimney extend, characterised by a significant increase (1.5 order of magnitude) in dynamic permeability.


Fluid flux through a horizontal slice of 1 m of low-permeable shale located at 1 m
above the source region showing corresponding typical circular craters or pockmarks.

Daniel Kiss | Spontaneous generation of ductile shear zones by thermal softening

Daniel Kiss | Spontaneous generation of ductile shear zones by thermal softening

The spontaneous generation of shear zones in ductile rock is fundamental for many geodynamic processes, such as the initiation of subduction, the generation of strike slip zones or the formation of tectonic nappes. Yet, it is a challenge to find a thermo-mechanically feasible mechanism to “break” (i.e. form localised ductile shear zones) the cold and hence strong parts of the lower crust and the lithospheric mantle. The conversion of dissipative work into heat, the related local temperature increase and the associated decrease of temperature dependent rock viscosities is one possibility.

As the video shows thermal softening can can result in spontaneous strain localisation: starting from a spherical seed (week inclusion) a planar temperature anomaly is developed. As a result two practically rigid blocks are formed with a ductile shear zone in between.

We are determining the physical conditions needed for such deformation mode, as well as the resulting shear zone sizes and temperatures.