Table of Contents
Workshop: Exascale I/O for Unstructured Grids (EIUG)
When dealing with large amounts of data, the I/O subsystem of modern Supercomputers is one of the most sensible components with regards to performance. Today, application scientists from different backgrounds have to struggle with the details of parallel file systems to obtain optimal performance. Only to a certain extend the management of the data is supported by libraries like HDF5. An especially challenging example is the storage of results computed on unstructured grids, maybe including grid adaptivity, occurring for example in computations in structural mechanics, cosmology or climate simulations.
The workshop will bring together experts from application development, middleware and storage design to discuss the issues of storing and accessing large datasets. While focusing on data for (unstructured) grids, the workshop generally addresses HPC data access. The main goal of the workshop is the identification of future proof strategies for the efficient access of large data sets with respect to an optimal utilization of modern storage subsystems and memory hierarchies.
It is intended to publish the results of the workshop in form of a white paper.
Date | September Monday 25th, Tuesday 26th, 2017 | ||
Venue | DKRZ, Hamburg, Germany | ||
Contact | Julian Kunkel |
Please join our mailing list.
This workshop is supported by SPPEXA, the DKRZ and powered by ESiWACE 1) and the Virtual Institute for I/O.
Relation to SPPEXA
The efficient handling of the parallel I/O of large datasets is a crucial problem for the operation of exascale computers. For large simulations datasets of Tera- to Petabytes have to been handled in one output operation, while a hole simulation run can produce up to Exabytes, which need to be managed effectively. Simulations which potentially produce large datasets, partly on unstructured grids, occur in a number of SPPEXA projects like EXA-DUNE, AIMES, Terra-Neo, EXASTEEL-2, ExaFSA, EXAHD, EXAMAG and ExaDG. There are also SPPEXA projects with focus on solutions for Exascale-I/O like ADA-FS and AIMES. The workshop will intensify contact between the SPPEXA and other experts from the area of parallel I/O, a relevant part of the HPC community, probably not completely aware of SPPEXA up to now. The workshop will also serve as a starting point for cooperation, e.g. common sessions on international conferences.
Organization
The workshop is organized by
- Julian Kunkel (DKRZ), kunkel@dkrz.de (AIMES project)
- Olaf Ippisch (TU Clausthal), olaf.ippisch@tu-clausthal.de (EXA-Dune project)
- Sebastian Oeste (TU Dresden), sebastian.oeste@tu-dresden.de (ADA-FS project)
Registration
To participate, please register online here: http://goo.gl/7LJyVR. Alternatively, you may send an email to kunkel@dkrz.de. Note that our seats are limited and we have to confirm your attendance (you may also kindly ask about the status).
Accommodation
Within walking distance, we recommend the: Mercure Hotel Hamburg Mitte
Address: Schroederstiftstr. 3, 20146 Hamburg - Rotherbaum, Deutschland
DKRZ has a customer ID that we can provide to you upon request; this should give you a condition of 79€ per night.
Please see the PDF for transportation from airport to the venue/hotel.
Agenda
The workshop is scheduled for a duration of two days, with extended discussion time between talks to find a common assessment of the challenges and possible solution strategies.
Monday
- 09:30 Coffee and Welcome
Julian Kunkel (DKRZ) – slides - 10:00 Evolving Sandia National Labs Engineering Unstructured Grids Codes for Exascale
Jay Lofstead (Sandia) – slides - 10:40 Parallel IO in the Met Office LFRic Infrastructure
Samantha Adams (Metoffice) – slides - 11:20 Scalable I/O for the earthquake simulation software SeisSol
Sebastian Rettenberger (TUM) – slides - 12:00 Lunch
- 13:00 IME: Freedom from Filesystem Foibles
James Coomer (DDN) – slides - 13:40 Campaign Storage
Peter Braam (Campaign Storage) – slides - 14:20 Looking into Clouds – Our approach for the visualization and analysis of large ICON data sets
Niklas Röber (DKRZ) – slides - 15:00 Break
- 15:30 I/O requirements for Exascale post-processing
Julien Jomier (Kitware) – slides - 16:10 Overview of file formats for (unstructured) data
Anastasiia Novikova (Uni Hamburg) – slides - 16:50 Discussion: Limitations of current systems
Discussion lead: Jay Lofstead (Sandia) – slides - 18:00 End
- 19:00 Social event – sponsored dinner in the Scandic Hamburg Emporio
Tuesday
- 08:55 Welcome from the director
Thomas Ludwig (DKRZ) - 09:00 I/O and Scheduling aspects in DEEP-EST
Norbert Eicker (FZ Jülich) – slides - 09:40 Reduction of data sets through high-level I/O interfaces
Yevhen Alforov (DKRZ) – slides - 10:00 Crowds and clouds - the potential of using data from smartphones in numerical weather prediction
Kasper Stener Hintz (University of Copenhagen) – slides - 10:20 Break
- 10:50 I/O Optimizations in ICON by the DWD
Nathanael Hübbe (Uni Hamburg, former DWD employee) – slides - 11:20 Practical I/O Analysis
Sebastian Oeste (TU Dresden) – slides - 12:00 Lunch
- 13:00 Large scale I/O benchmarking for structured grids and implications for unstructured grids
Olaf Ippisch (TU Clausthal) – slides - 13:40 Space and Performance Optimizations of NetCDF-Grids
Eugen Betke (DKRZ) – slides - 14:20 Towards new storage interfaces – chance or curse?
Julian Kunkel (DKRZ) – slides - 15:00 Discussion: Solutions, emerging questions
Julian Kunkel – slides - 16:00 Farewell