ExaNLA

Exascale-ready distributed and accelerated dense Numerical Linear Algebra

ExaNLA is a brainstorming workshop named after a project aiming at creating an innovative numerical linear algebra library for selected application-inspired kernels capable of executing at scale on the JUPITER Booster and similarly massively parallel accelerated architectures. 

Place: Rotunda hall and Foyer, Juelich Supercomputing Centre, Forschungszentrum Juelich

Date: February 11 2025 08:30 - February 13 2025 14:00

Event page: https://indico3-jsc.fz-juelich.de/e/exanla

The advent of JUPITER, an upcoming exascale supercomputer based on NVIDIA accelerated technology, is making the need for leadership in numerical algebra software. JUPITER’s introduction highlights the existing deficiency in distributed dense Numerical Linear Algebra libraries (NLA) that can effectively harness its enormous computational power. As JSC prepares to integrate this exascale supercomputer into its resources, there is a critical need for software solutions that can manage and optimize computations at this scale. The lack of a clear candidate for such NLA library amounts to an entry-point barrier to fully utilizing JUPITER’s capabilities for scientific research and industrial applications, where distributed numerical linear algebra is essential for solving complex problems.

To address the need for advanced NLA software that can fully exploit the capabilities of JUPITER and future exascale supercomputers while ensuring robust fault tolerance and checkpointing, a comprehensive strategy must be developed. Realizing a new numerical linear algebra library targeting large distributed and accelerated supercomputing platforms is not a simple feat. It requires a careful and detailed plan and it needs to involve key scientists and developers in numerical algebra as well as in the application codes that would benefit from such new library. For these reasons we organize a workshop with a number of goals general enough to allow for adjustments in due course, and at the same time specific enough to ensure success.

Objectives

  1. Engage with the Community of Expert Users: Conduct an in-depth consultation process with domain experts to identify the essential core functionalities required by application codes, ensuring that the ExaNLA library meets the most pressing needs of its primary users.
  2. Initiate Brainstorming Sessions: Begin a series of collaborative discussions to address the needs and challenges in developing the new ExaNLA library.
  3. Designing for Performance and Adaptability: Explore the feasibility of incorporating automatic mechanisms within the library to enhance performance and adaptability.
  4. Establish a Robust RSE Framework: Set up a research software engineering (RSE) framework based on Agile methodologies to ensure a streamlined and efficient development process. 
  5. Application-driven benchmarking: Perpetual execution of application-driven benchmarks, including comparison against evolving existing libraries.

Program

The workshop is divided in 4 sessions spread over two and half days. Each session is further divided reflecting the objectives that the workshop wants to achieve. Each part of the workshop lasts 1.5 hours and is opened by a short talk (~20 minutes) given by an expert which sets the stage for that particular session. Every participants is free to contribute to each session with a 5-10 minutes (2-5 slides max.) presentation or to bring in specific challenges and insights that he/she deems worth discussing. Here is the full list of sessions:

  • Session 1: Application-driven functionalities
    • Survey of functionalities
    • Selection of kernels
  • Session 2: Technical brainstorming
    • Programming languages
    • Parallel programming models
    • Interfaces and templating
    • Fault tolerance
  • Session 3: Designing for performance and adaptability
    • Automatic symmetry discovery and algorithm selection
    • Architecture-driven optimizations
  • Session 4: Software development and benchmarking
    • CI/CD and Test-driven development
    • Continuous and application-driven benchmarking

PARTICIPATION
Attendance to the event is by invitation only. The registration is moderated and it will be confirmed shortly after submission. It is expected to have participants from several disciplines spanning from application fields (e.g., physics, engineering, chemistry), to computer scientists, parallel programming experts, applied mathematicians and software engineers. To register to the workshop please use the link below and make sure to fill all pertinent fields in the form
Registration

Organization

The event is organized by the Simulation and Data Lab Quantum Materials which is part of the division High-Performance Computing for Quantum Systems of the Juelich Supercomputing Centre. All meals and coffe break during the workshop are offered by the Juelich Supercomputing centre. As part of the workshop there will be a Social Event the evening of February 12. Details will be provided during the workshop. Traveling and accommodation cannot be provided by the organizing host and will be exclusive responsibility of the participants. For a list of possible accommodations and Hotel please see the incomplete list below.

Accommodation

Juelich is not a large city but there are several Hotel routinely used by visitors of the Forschungszentrum Juelich. They are listed below with a link to their website

Transport

The Forschungszentrum Juelich can be reached by public transportation by train and/or bus. Visitors can also access the research center with a private vehicle. External visitors need to register at the visitor center. We recommend that all participants external to the Forschungszentrum fill up the appropriate fields in the registration form. This will allow the organizing committee to pre-register external participants which will be provided with a pass lasting the full duration of the event without the need to repeat the registration each day. The organizing committee will try to organize a shuttle service from/to Juelich town center for the participants lodging in the Hotels in town.

The Jülich Supercomputing Centre (JSC) is located within the Forschunhgszentrum Jülich (FZJ). FZJ can be reached by car or by a combination of Deutsche Bahn train, Ruhrtalbahn and local bus (219). Due to its central location in the German state of NRW, the two most convenient international airport are:

Last Modified: 22.01.2025