By Suchismita Sanyal, Shell’s General Manager for Computational Science. on Nov 6, 2020
Shell maintains a top-tier HPC facility to enable faster and sharper insights from computational sciences, aided by advanced visualisation. The technology is used across Shell’s businesses to analyse data and understand phenomena in everything from Earth sciences to the chemistry of catalysts and batteries as well as the flow of liquids and gases through reactors, pipelines and rocks.
HPC enables running data driven algorithms that are dedicated to solving Shell’s most challenging business opportunities, faster, and for real-sized systems. Simulations are increasingly replacing laboratory experiments and even whole pilot plants. The main HPC areas of application for Shell are geophysics, seismic imaging, material science, and fluid dynamics at various scales. Our growing businesses in low-carbon energy solutions are beginning to rely on HPC as well.
Digitalisation brings more data, which in turn allows greater insights for the businesses. Accurate simulations are critical to understand complex phenomena and usually involve solving numerical algorithms with high computational complexity. To be able to make real-time decisions on this data, it is critical to optimise the efficiency of the computer systems on which these simulations are performed.
This is where HPC systems, also known as “supercomputers", come into play. They solve computational problems that are either too large for standard desktop computers alone or would take too long to solve. An HPC system is essentially a network of many computers, each of which contains heterogeneous processing units like central processing units (CPUs) and graphic processing units (GPUs). Hundreds of Shell end-users across 18 countries are able to perform calculations around the clock thanks to our network of datacentres.
“Supercomputers allow us to leverage the avalanche of available operational data into added value for our businesses.”Detlef Hohl, Shell's chief scientist for computational and data science
Shell’s Computational Science Centre of Excellence has its stronghold at the Shell Technology Centre Bangalore in India. There, computational science researchers develop, deliver and use application software for HPC systems.
An example of HPC in our conventional operations is the estimation of rock properties using computer simulations. Rock core sampling is utilized to determine the physical and chemical nature of the rock of a hydrocarbon reservoir. Digital modelling of the reservoir’s rock based on the computed tomography scanning (CT scan) of a core sample is now being used to accelerate the evaluation of hydrocarbon saturation levels, permeability and porosity of a reservoir’s rocks. Here, reaching high-resolution, trustworthy results from simulations requires significant computing power. Our scientists recently optimized and tailored an image reconstruction algorithm to satisfy requirements for speed of execution and high quality of images required for business decision making.
The original algorithm came from the medical imaging technique of computed tomography scanning (CT scan). The code was transposed from the MATLAB coding language into C++ code and then adapted to run on a Compute Unified Device Architecture (CUDA). The original algorithm took about 5 hours to run on 8 GPUs. After the team’s optimisations, it can run in just 2 minutes on a single GPU. The optimisations allows the code to run iteratively at affordable costs. That means, that the process can be repeated as often as needed until it generates the required outcomes. The unique insights derived from such HPC usage drives more value out of each dollar invested in the discovery of new resources.
Another successful code-optimisation was recently achieved thanks to our strong collaboration with NVIDIA. Shell has researched flow properties inside rock using physics based mathematical modelling since the 1960s. Digital Rock modelling accounts for around 10% of Shell HPC usage. For this application, whose results are published here, our scientists made the existing Shell proprietary code faster using advanced HPC technologies that reduce the time needed to ingest data and run algorithms.
Our collaboration with NVIDIA was an important aspect of this project because it brought about significant changes in the algorithms and source code. By working together, NVIDIA engineers and Shell’s scientists improved the execution time of this algorithms up to eightfold. Simulations which took more than a month to run can now be executed in four days. Collaborating with others is an essential aspect of our strategy to bring in cutting-edge solutions and improve our use of HPC.
By combining our in-house expertise with other technology giants, we are able to reach new heights and gain entirely new insights for our business.
At Shell, computational science brings technology improvements that push the boundaries and contribute to Shell’s transformation into a net-zero emissions energy business. Our HPC infrastructure and code optimisation capability have proven to be true differentiators in helping us accelerate the energy transition.
Suchismita Sanyal is the General Manager, Computational Science at Shell Technology Centre Bangalore. She leads a group of 60 researchers in the Computational Science group in Shell, delivering digital solutions across multiple Shell businesses and designed to position Shell on a stronger foothold in the Energy Transition.