Access Abstraction to HPC Resources - Virtual Event

Akademie I (Empa)

Akademie I


Überlandstrasse 129 8600 Dübendorf
Carlo Pignedoli (Empa), Daniele Passerone (Empa), Edoardo Baldi (Empa), Michele De Lorenzi (ETH Zurich / CSCS)


For HPC centers it is essential to simplify the access to the computational resources for their user  communities. Personal computing did already introduce in the eighties of last century the concept of desktop computing to simplify the utilization by the users. At the same time we moved from mainframes to client-server architectures up to micro-services. Unfortunately supercomputers are still (mostly) used by manually submitting jobs in queues.

Some scientific communities did already work on such challenges and created web portals able to support complex workflow hiding from the end users details related to the HPC infrastructure. More and more scientists will be involved in the next few years in building such platforms.

For us it is getting more and more important to support those portals and offer new ways to use the computational and storage resources we manage. It shall be possible to use the supercomputers interactively or through defined programmatic interfaces (API) to cover from simple use cases up to complex workflows. This also includes access to data management tools to stage in (and out) everything that’s needed to run the simulation or analysis.

Key Questions

  • Scientific portals: What kind of scientific portals do exist today? What will be the future developments? What are their requirements respect to the HPC providers?

  • Interactive computing platforms: How can I easily run a calculation/analysis that requires some degree of interaction on a bigger and more powerful machine? Can I use an HPC resource as if it was my laptop?

  • Efficient management of data: Are there any optimal tools to manage data efficiently? E.g., avoiding copying non-essential info when moving data, or saving data for archiving purposes  Can I synchronize files between my computer, my home institution and my HPC resource without having to use too many different tools? Without having to monitor failures manually?

  • Computational resources abstraction: As a HPC provider, how can we build a “middle layer” allowing our scientific computing users to see the resources we are providing in a uniform way? (keywords: workflow, API, RESTful interfaces).

  • Adam Henderson
  • Alain Studer
  • Alexander Kashev
  • Alexandre Wetzel
  • Alfio Lazzaro
  • Allen Neeser
  • Andrei Plamada
  • Anibal Moreno
  • Antonio Russo
  • Carlo Antonio Pignedoli
  • Carlos Fenoy
  • Christian Bolliger
  • Daniele Passerone
  • Edoardo Baldi
  • Eirini Koutsaniti
  • Elmar Dette
  • Enrico Favero
  • Filippo Stenico
  • Fotis Georgatos
  • Gianfranco Sciacca
  • Giovanni Pizzi
  • Guillermo Losilla
  • Hans-Rudolf Hotz
  • Hardik Kothari
  • Heinrich Billich
  • Jean-Baptiste Aubort
  • Jean-Claude De Giorgi
  • Kristjan Eimre
  • Leonardo Sala
  • Maria Grazia Giuffreda
  • Martin Jacquot
  • Massimo Brero
  • Maxime Martinasso
  • Mei-Chih Chang
  • Michael Niedermann
  • Michele De Lorenzi
  • Nicolas Kowenski
  • Olivier Byrde
  • Pablo Fernandez
  • Pascal Häussler
  • Patrick Bleiziffer
  • Patrick Zosso
  • Radim Janalík
  • Raluca Hodoroaba
  • Rene Windiks
  • Ricardo Silva
  • Roberto Fabbretti
  • Rok Roškar
  • Sean Hughes
  • Sofiane Sarni
  • Stefano Gorini
  • Sébastien Moretti
  • Thomas Kramer
  • Thomas Wüst
  • Tim Robinson
  • Ulrich Tehrani
  • Victor Holanda
    • 10:00 10:15
      Welcome and Introduction
      Conveners: Daniele Passerone (Empa), Michele De Lorenzi (CSCS)
    • 10:15 10:45
      Reproducible High-Throughput HPC Workloads with AiiDA 30m

      The ever-growing availability of computing power and the sustained development of advanced computational methods present new challenges driven by the sheer amount of calculations and data to manage. In this talk I will present AiiDA (, a robust open-source high-throughput infrastructure addressing the challenges arising from the needs of automated workflow management and data provenance recording. I will discuss the developments and capabilities required to reach sustained performance, with AiiDA supporting throughputs of tens of thousands processes/hour, each of them ranging from local processes to massive HPC jobs. I will also discuss how AiiDA automatically preserves and stores the full data provenance in a relational database making it queryable and traversable, thus enabling high-performance data analytics and reproducibility of the research.

      Speaker: Giovanni Pizzi (EPFL)
    • 10:45 11:15
      AiiDA Lab – an Ecosystem for Developing, Executing and Sharing Scientific Workflows 30m

      Computational simulations constitute an integral part of modern science. Running scientific software, however, requires expert knowledge not only in the modelled scientific phenomena but often also in accessing and operating compute resources, programming, installing and compiling software, and other topics related to computation. In the presentation I will introduce AiiDA lab, an open source platform that provides the means to develop, execute and share computational science workflows as an intuitive web service available to anyone who knows how to use a web browser. AiiDA lab is built on top of AiiDA, an infrastructure that provides the automatic workflow engine and provenance tracking, and Jupyter, which powers the graphical user interfaces.

      Speaker: Kristjan Eimre (Empa)
    • 11:15 11:45
      FirecREST: a RESTful API to HPC Systems 30m

      As science gateways are becoming an increasingly popular digital interface for scientific communities, it is important for High-Performance Computing centers to provide a modern and externally accessible interface such as Web-enabled APIs. Such interface provides access to HPC center resources to allow scientific web portals to submit jobs or move data in and out the HPC center. This work presents the FirecREST API, a RESTful Web API infrastructure that allows scientific communities to access the various integrated resources and services available on HPC systems.

      Speaker: Eirini Koutsaniti (CSCS)
    • 11:45 12:15
      There is no Silver Bullet - Improved HPC User Experience for Different User Groups
      Convener: Christian Bolliger (ETH Zurich)
    • 12:15 13:15
      Lunch 1h
    • 13:15 13:45
      Real-Time Services for Large Volume Experiment-Data Analysis Utilizing Supercomputing and Cloud Technologies (SELVEDAS) 30m

      The ongoing developments in accelerators, detectors and experiment automation is leading to a rapid growth of data generated during experiments. A viable solution is utilizing suitable infrastructures that allow additional remote high performance capacity for processing and analysis of data from the experimental facilities with larger data volumes and higher processing needs.

      The SELVEDAS project proposes a hybrid cloud infrastructure, offering scalable and extensible services for data management and analysis to Swiss academic users by leveraging high performance computing (HPC), storage, networking as well as cloud technologies and orchestration. The on-demand services perform as a highly efficient remote data processing system providing fast feedback and analysis with the long time storage and archival of petabytes of data.

      Speaker: Mei-Chih Chang (PSI)
    • 13:45 14:15
      Interactive HPC on Piz Daint with JupyterLab 30m

      The last few months have seen a significant rise in the number of users accessing Piz Daint interactively via Jupyter notebooks, and we expect this trend to increase further in the future. We discuss some of the challenges faced in providing interactive computing in a multi-user, batch-oriented computing environment such as Piz Daint. We describe our centrally-maintained JupyterLab software stack and some of the tools we provide to enable customized workflows.

      Speaker: Maxime Martinasso (CSCS)
    • 14:15 14:20
      Farewell and End of the Meeting