The UK’s MMM community have come together to establish which areas of science we want to explore once Exascale HPC arrives, the impact that completing these simulations would have, and who in the UK is actively involved in this endeavour. The overriding aim is to develop world leading software, suitable for high-impact scientific applications on heterogeneous Exascale computing hardware.
The range of application ranges from astrophysics and cosmology, through engineering fluid flows to molecular dynamics and the development of novel materials. These aims will be met through a linked set of 4 work packages:
WP 1 Improving the task scheduling system.
WP 2 has two parts
2.1 Exploiting heterogeneous architectures and novel network hardware. The objectives are:
* Develop task kernels to execute tasks on a wide variety of architectures including different GPU accelerators
* Work with cross-cutting groups to develop flexible ways to define kernels and work between different architectures, making use of generic languages such as SYCL and OpenACC
* Work with hardware vendors to co-design novel and flexible solutions to communication delays between domains
* Develop ways of exploiting direct communication between GPU systems
2.2 Developing algorithms that are well-suited to the task system and heterogeneous architectures. The objectives are:
* Develop optimal low-level algorithms, common to many science problems
* Make optimal use of accelerator hardware for complex multi-centre interactions
* Design algorithms for acceleration of DFT and quantum chemistry codes
WP 3 focuses on the complex workflows required for meta-level application of the simulations, e.g. material discovery or interactions between coupled codes. The objectives are:
* Develop interfaces to code and data to allow a wide range of codes to be brought together in a single workflow
* Develop minimisation techniques and exploit the workflow for template problems, such as material discovery
* Develop workflows to exploit the performance of GPU-enabled code
* Couple codes within the task system to simplify the implementation of multiphysics problems
WP 4 focuses on knowledge exchange activities to
* Develop a community of expertise to leverage developments across science domain boundaries
* Engage the science community in the project
* Outreach, including to schools and the general public
* Foster close working relationships with hardware vendors and software standards groups
* Develop international collaborations and exchange of expertise
Our proposal and work packages address the four pillars of Excalibur by, amongst others, novel uses of Taskification; novel algorithm and data decomposition investigation and development; novel workflows and workflows management systems; and novel interactions between apparently disparate scientific communities.