The aim of “PRACE-5IP WP7: Applications Enabling and Support” is to support European researchers to ensure that there are a broad range of important applications that can run efficiently on state-of-the-art supercomputers.
The objectives for WP7 in PRACE-5IP include:
- Applications Enabling Services for PRACE Systems
- Preparing for PRACE Exascale Systems
- Supporting European HPC Researchers
- Provision of Numerical Libraries for Heterogeneous/Hybrid Architectures
These build on the previous successful applications activities in PRACE, with an enhanced focus on Exascale and a clear intention to further strengthen collaborations with CoEs and FET-HPC projects.
In this page please find a summary of PRACE-5IP WP7 Activities with links where more information can be found.
Summary of PRACE-5IP WP7 Activities
Task 7.1: Applications Enabling Services for PRACE Systems
Task 7.1 is comprised of the following two subtasks.
T7.1.A: Enabling Applications Codes for PRACE Systems – Preparatory Access Type C&D
PRACE offers the possibility to test and optimise scientific applications through the PRACE Preparatory Access calls to prepare an application for a regular PRACE project access proposal. In addition to granting computing time on a PRACE system, Preparatory Access Type C and D also allows researchers to ask for collaboration with PRACE experts from the PRACE project. The expert helps in the context of optimization and profiling work and provide HPC knowledge.
Type C allows direct optimization work on a Tier-0 system while Type D allows researchers to start the optimization work on a Tier-1 system to finally reach Tier-0 scalability.
Within the last three years, 24 projects were or are currently supported by PRACE experts. Most of them also published a white paper covering the technical optimization work. These can be found at: http:/
More details about the PRACE Preparatory Access calls and how to apply can be found at: http:/
SHAPE, the SME HPC Adoption Programme in Europe is a pan-European, is a PRACE-based programme supporting HPC adoption by SMEs. A series of regular calls offer European SMEs the chance to get assistance from PRACE to realise the potential of HPC within their organisation. Successful SHAPE projects typically receive 3-6 months of effort from a PRACE expert to assist with porting or setting up their codes to work in a HPC environment.
Since 2013, over 40 SMEs have received SHAPE effort across the fields of computational fluids dynamics, steel casting, medicine, genomics, environment and renewable energies and artificial intelligence. As a result, jobs have been created, costs reduced, contracts won and new and innovative services have been offered. Calls will continue to be opened regularly offering further SMEs the opportunity to participate.
More information about SHAPE can be found at: http:/
T7.2: Preparing for PRACE Exascale Systems
The objective of T7.2 is to facilitate preparing the techniques, tools and application for upcoming European exascale systems. This is achieved by investigating the tools and techniques developed in European projects and applying them to important applications that are widely used by the academic and industrial communities. The task runs between January 2017 and March 2019.
The objectives of T7.2 are achieved by implementing 8 mini-projects within the task, each of which is an engagement between FETHPC projects and Centres of Excellence. The task is implemented in the following two phases:
Phase 1: This was a survey phase implemented between January 2017 and January 2018 where a number of workshops were held to facilitate communication between FETHPC projects and CoEs. The result of this phase is a set of 8 mini-projects that were reviewed and approved by a committee formed by the PMO. Each mini-project is executed by a PRACE partner, a FETHPC tool/technique and an application from a CoE.
Phase 2: This is the ongoing exploitation phase between February 2018 and March 2019 where the implementation of all the mini-projects is ongoing. The first step of this phase for each mini-project was to define a workplan to implement the mini-project between March 2018 and March 2019. As of December 2018, all mini-projects are on going according to their individual workplans. The results from each mini-project will be summarised in a white paper by the PRACE, FETHPC and CoE partners, and their summaries included in deliverable “D7.3 – Evaluation of Tools and Techniques for Future Exascale Systems” in March 2019.
The status and progress of T7.2 were presented at the following events during the exploitation phase:
- Invited talk at the PRACE booth at ISC, June 2018.
- PRACE-CoEs-FETHPC-EXDCI Workshop, October 2018.
- Links to events and talks related to T7.2:
Task 7.3: Supporting European HPC Researchers
Task 7.3 is comprised of the following two subtasks.
This activity of this task is to update and maintain the Unified European Application Benchmark Suite (UEABS) so that it can be used in future procurements and to help European researchers choose systems that are appropriate for their computational requirements. The UEABS and the Accelerator Benchmark Suite have been merged to a newly re-unified UEABS 2.0 and migrated to the PRACE GitLab server: https:/
We will evaluate benchmark results on: the PRACE Tier-0 systems, the PCP Prototypes, the DEEP-ER SDV system and the Mont-Blanc 3 Dibona system. We will publish comparative results and analysis including performance scaling and energy efficiency (where possible) in deliverable “D7.5 – Evaluation of Accelerated and Non-accelerated Benchmarks”. We will also publish the results in the PRACE GitLab UEABS repository.
T7.3.B: Best Practice Guides
WP7 publishes the successful series of Best Practice Guides on new architectures and systems. Topics covered include system architecture and configuration, production and programming environment, performance analysis, tuning and debugging. Some of the Best Practice Guides published by PRACE belong to the most frequently accessed documents on the PRACE webserver. We plan to publish around 2 new Best Practice Guides per year.
Likely topics of technical interest include new processors or GPUs, new memory technologies (MCDRAM, NVRAM), new interconnects and workflows for HPC job processing and data management.
Further information can be found at: http:/
Task 7.4: Provision of Numerical Libraries for Heterogeneous/Hybrid Architectures
The activity of task 7.4 is focused on the development of a standardised API allowing the use of different implementations of underlying numerical libraries, optimised on different computing architectures. The standardised API will enable developers of scientific applications to adopt the best implementation of a library in a seamless way, without time consuming changes in the source code. ETHZ is leading the effort in task 7.4 with the contributions of CINECA, NCSA, CaSToRC: the application areas of task 7.4 are Linear Algebra and Molecular Dynamics.
Task 7.4 has been able to deploy the main infrastructure of the linear algebra library, with the matrix class and communicator grid creation and Scalapack to tile distribution conversion. Libraries currently supported are Scalapack (SL), DPlasma (DP, CPU only) and ELPA. Routines currently supported by C++ and C/Fortran interface are Cholesky factorization (SL, DP), matrix matrix multiplication (SL, DP), limited LU factorization (SL, DP) and eigenvalue solver (SL, ELPA). The unified API for non-bonded interactions in classical molecular dynamics is going to be merged in GROMACS, thanks to the involvement of the GROMACS development team. We have finalized the standardised API for the performance counters and how to expose it to the user, introducing the schedule entity as well, which controls the scheduling of non-bonded kernel execution. The partner NCSA developed a library for long-range electrostatic interactions, with a direct Poisson’s Equation solver based on Pseudo-Spectral Method and Periodic Boundary Condition. CINECA has released a set of autonomous mini-apps to perform scientifically relevant linear algebra operations: the software are LAXlib miniapp generating a random hermitian matrix and performing parallel diagonalization, miniapps reading matrices extracted from real input cases, available in Quantum ESPRESSO GitLab.
WP7.4 plans to fix the DPlasma issue compiled with GPU support for the currently supported libraries and to extend the supported routines including the triangular system and the matrix inversion. The next steps will be porting the Schedule entity into GROMACS and test its performance, introducing cleaner data structures for input and output for the API (currently we are passing around raw pointers in order to accommodate the implementation of non-bonded interactions). NCSA plans to implement short-range interactions kernel with scalability optimizations and validation. CaSToRC tried to adapt LU-decomposition locally in the Lattice QCD code, but it did not improve the methods currently in use. A possible implementation of eigenvalue solvers is still under discussion.
We have summarized the information on WP7.4 in the web page: https:/