Materials and Molecular Modelling Exascale Design and Development Working Group

点击翻译按钮获取中文摘要

High Performance Computers (HPC), or supercomputers, offer exciting opportunities in understanding, developing and increasingly predicting the properties of complex materials through atomistic and electronic structure modelling; and the scope and power of our computational techniques continue to expand as the capability of the hardware grows. The advent of exascale systems is the next dramatic step in this evolution. There is a high cost of both purchasing and running such a system, so it is imperative that appropriate software is developed before users gain access to exascale facilities. The investigators of this project are internationally leading experts in developing (enabling new science) and optimising (making simulations more efficient) state-of-the-art materials software for running simulations on HPC, based here and abroad. Software that we have developed is used both in academia and in industry. Currently, our community consumes over a third of the UK's HPC facility (ARCHER) of approximately 120,000 cores (with a peak performance of 2.5x10^15 Flop/s) and we do not anticipate any delays in immediately getting the most out of the successor to ARCHER, which will be composed of approximately 750,000 cores (estimated at ~ 28x10^15 Flop/s) when it becomes available later this year to the UK academic community. Exascale computers (by definition >10^18 Flop/s) will be composed of many more cores. However, it is anticipated that the next generations of national computers will not provide a smooth transition from the existing infrastructures, but instead will undergo a step change for the UK national facilities, with a shift from conventional CPU based architectures to CPUs hosting (multiple) many-core accelerators. Many, if not all, of our software packages will require major changes before these architectures can be fully exploited. Appropriate changes to the software will effect a reduction in data exchange between cores, management of communications between CPU and accelerators, and, moreover, adaptation in our procedures for handling input and output data.As requested in the Call, we will form a design and development working group (DDWG) by bringing together Research Software Engineers (RSEs) and experts from mathematics and computer science with a wide range of domain experts in Materials and Molecular Modelling (MMM). We represent a very large and important community which has long established mechanisms for development, resource management, and disseminating best practices that the DDWG will exploit. To maximise the impact of the available UK funds for exascale computing, we identify solutions that will benefit most of our community.Our DDWG aims to separate out the fundamental mathematics of the problem from the computer science of implementation. We will exploit best current practices and those under development in our domain and across other disciplines in particular targeting libraries that can be called by many materials software and offer a route to heterogeneous architectures. In a complementary development, we will tackle new workflows to manage and analyse vast volumes of simulation data. We will gain valuable experience from our regular meetings and meetings with other DDWGs; and knowledge transfer with our national and international project partners. Undertaking of the initial work and identifying what is required (work earmarked as part of the next funding stage) will enrich our expertise and facilitate international collaborations with developers of materials software and users of overseas exascale computers.This work enables the UK MMM community to use exascale HPC resources efficiently to address many EPSRC Grand Challenges, including Emergence and Nanoscale Design of Functional Materials (Physics); Dial a Molecule and Directed Assembly of Extended Structures with Targeted Properties (Chemical Sciences); and Engineering From Atoms to Applications (Engineering).