Lattice Field Theory at the Exascale Frontier
Lead Research Organisation:
Swansea University
Department Name: College of Science
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
Organisations
People |
ORCID iD |
| Biago Lucini (Principal Investigator) |
| Description | Exascale computers will soon be available to the scientific community for research purposes. This new generation of hardware will allow new, ambitious scientific objectives to be achieved provided that the practitioners in the field of High-Performance Computing developed the knowledge and the software to use effectively the new architectures. New programming paradigms need to be developed and implemented effectively. The Exalat project has laid the foundations for the design of new algorithms and their benchmarks on existing supercomputers, while simultaneously surveying the scientific opportunities for the Lattice Field Theory community in the UK, and establishing an effective training pattern for all practitioners in the field, academics, students, PDRAs and RSEs. We are currently in the process of finalising a set of partnership activities with companies active in the HPC sector. This grant is part of the Excalibur-funded high-priority use case EXALAT, led by Edinburgh. Those reported in this section are the findings of the collaborative project, in which Swansea has participated as a partner. |
| Exploitation Route | The Exalat project has set up activities in four main areas: benchmarking, algorithm development, training, and data curation. Workshops have been organised where the activities were discussed and results presented. The response from the UK community has been positive, with a seemingly high level of engagement. All the outcomes of these activities are public and available to the whole community, in particular the training material. most of which have been developed especially for Exalat. It will be collected on our website and will build in time a unique collection of material produced by practitioners dedicated to starting joint activities with other Excalibur work packages. We plan to continue these activities regularly in the coming years. Indeed, the activity of Exalat has generated a new initiative, ExaTEPP, which is a collaborative effort with another Excalibur-funded high-priority use case. The legacy of Exalat is informing and influencing the initiatives and the progress of ExaTEPP. |
| Sectors | Digital/Communication/Information Technologies (including Software),Education |
| URL | https://www.exalat.org |
| Description | The main non-academic impact of High-Performance Computing projects like this one comes from the training of a new generation of practitioners who are capable of developing software on the latest architectures available on the market. Through our training events, we have had the opportunity to train approximately 100 people during the lifetime of the project. Lattice Field Theor has a long track record of producing computer-literate graduates who then work in multiple sectors of the economy (digital. finance, education). The work done by the exalat project has been influential in developing and delivering the benchmark of a new DIKAC HPC system at Edinburgh. The benchmark routines developed in Exalat have been contributed to Excalibur and used in the RSE working group. Our benchmark suite has been used by leading High-Performance Computing companies and research centres to assess the capabilities of their systems. |
| First Year Of Impact | 2021 |
| Sector | Digital/Communication/Information Technologies (including Software) |
| Impact Types | Economic |
| Title | SOMBRERO Benchmarking Suite |
| Description | SOMBRERO is a suite of benchmarks for High-Performance Computing systems. SOMBRERO is derived from HiRep, a research code used in theoretical particle physics for studying strong interactions beyond the standard model. The physical properties of the interactions simulated by HiRep give SOMBRERO a unique ability to vary the ratio between computation and communication by simply changing the internal degrees of freedom of the physical model. This ability makes SOMBRERO an ideal tool for exploring performance on supercomputing systems under workloads of varying intensity. |
| Type Of Technology | Software |
| Year Produced | 2021 |
| Open Source License? | Yes |
| Impact | SOMBRERO has become part of the UKRI ExCALIBUR benchmarking suite and has in fact provided a template for developing other benchmarks. SOMBRERO has been featured in the Linux Format Magazine, which targets a wide lay audience. SOMBRERO is being adopted by leading HPC companies to assess the performance of their systems. |
| Description | Intel Hackathon and training for Research Software Engineers |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Professional Practitioners |
| Results and Impact | Code portability and avoidance of lock-in into proprietary specifications are strong principles of software development aimed at targeting future exascale systems, which are going to be based on a range of heterogeneous architectures. The ability to migrate seamlessly code from one architecture to the other will future-proof any program currently being developed. This aspect has been recognised also by vendors, who are abandoning the model of proprietary libraries in favour of multi-architecture open standard specifications. Among them, OpenMP, which is backed by a range of HPC vendors, is attracting increasing attention. Originally born as a set of libraries and directives for simplifying programming on shared memory architectures, OpenMP has now been extended, with the release of version 5 of the standard, to manage off-loading of the code to hardware accelerators such as GPU, enabling programmers to bypass lower level and architecture-specific instructions. In collaboration with Intel, EXALAT has organised a three-day Hackathon including seven hours of work on codes and two lectures covering OpenMP 5 and Data Parallel C++, a more comprehensive set of instructions that provide an alternative approach to portability through the increasingly popular SYCL standard. The event has been attended by sixteen members of the community, and has been supported by two EXALAT Research Software Engineers and five Intel staff. Participants have worked at improving the Grid and OpenQCD codes. In the case of Grid, the staggered kernel has been optimised, bringing its performance to levels that are theoretically comparable to those already obtained by the Domain Wall and by the Wilson kernel. Improvements have been made also to the OpenQCD code, for which crucial routines have benefited from a significant speed up. Before the hackathon, Intel has delivered four hours of training to EXALAT RSEs, to enable them to transfer the acquired knowledge to the project and to the community. This advanced training event has been joined also by five leading community software developers. The RSE training has covered the following points: • Introduction to OpenMP offloading in the Intel oneAPI development kit and case study of the NWChem software • Introduction to the Intel DevCloud, including demonstrating an example of the ISO3DFD code • Introduction to Intel Advisor for offload modelling • Profiling GPU offloaded code with the Intel VTune profiler |
| Year(s) Of Engagement Activity | 2021 |
| Description | Workshop on benchmarking |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Benchmarking is one of the four strands that in the EXALAT inaugural workshop have emerged as strategically very important for Lattice Gauge Theory towards the exascale. The workshop on benchmarking has taken place on Tue 8th of September, 2 - 4 UK time. The purpose of the workshop was to start to develop a roadmap for benchmarking, working towards the following objectives: (1) Identify a set of standard codes that are relevant for the Lattice Field Theory Programme and extract from them representative benchmarks that will be maintained as a part of the project (2) Identify figures of merit and key performance indicators on current architectures that can be used to characterise performance on the upcoming architectures (3) Adapt the benchmarks to the most relevant architectures, as they become available (4) Ensure that the benchmarks remain relevant by implementing all the algorithm improvements that become available in the lifetime of the project Part of the workshop has been devoted to short presentations on the meaning of benchmarking for our Lattice and on the current Lattice benchmark related to codes used in the UK. The presentations were then followed by broad discussions that led to the creation of a roadmap for Lattice benchmarking. |
| Year(s) Of Engagement Activity | 2020 |
| URL | https://www.exalat.org/2020/08/27/exalat-benchmarking-meeting-8th-of-september-2-4-bst/ |