1st Block: Parallel I/O and portable data formats (HDF5 and NetCDF)
Date and Time: Monday, October 19th, 2015, 09:00 - 17:00
Location: TU Wien, Festsaal, Karlsplatz 13, 1st floor
Lecturer: Sebastian Lührs, Jülich Supercomputing Centre
Details and Registration see: http://training.vsc.ac.at/course/index.php?categoryid=4
Abstract:
Processing large data sets on high-performance computing (HPC) systems including reading input data and writing results faces an ever growing need for scalability. In addition, input and output data are often processed on different, heterogeneous platforms. Conversion processes tend to be hard to maintain and can be prone to errors. Portable, self-describing data formats such as HDF5 and netCDF can help to solve these problems. They are widely used in scientific computing across multiple disciplines. This course will give an overview on the basics of HPC parallel I/O and introduces to the use of the HDF5 and netCDF library interfaces on HPC systems.
2nd Block: Node-level performance engineering: hardware and software aspects
Date and Time: Friday, December 11th, 2015, 13:00 - 17:00
Location: TU Wien, FH Hörsaal 2, Wiedner Hauptstraße 8-10, 2nd floor, yellow area
Lecturers: Karl Rupp and Josef Weinbub, Institute for Microelectronics, TU Wien
Registration: If you would like to join for this event, please email to: vsc-seminar@list.tuwien.ac.at
Abstract:
Learn about the fundamental principles of performance engineering for high performance computing: We introduce you to important hardware aspects, explain ccNUMA and demonstrate its pitfalls, and look into the inner workings of caches. After mastering the hardware fundamentals, we move on to the art of writing efficient software and discuss tradeoffs of different programming models and languages, in particular MPI and OpenMP. By the end of this tutorial you will have a deeper understanding of CPUs, GPUs, and other multi-core architectures such as Intel's Many Integrated Core (MIC) architecture, know how to derive performance models for your particular application, and have a richer toolbox of programming tricks and performance-optimized libraries to write efficient software to better answer challenging scientific questions.
3rd Block: Effective MPI programming: concepts, advanced features, dos and don'ts
Date and Time: Wednesday, January 13th, 2016, 13:45 - 17:45
Location: TU Wien, FH Hörsaal 7, Wiedner Hauptstraße 8-10, 2nd floor, yellow area
Lecturer: Jesper Larsson Träff, Research Group Parallel Computing, Institute of Information Systems, TU Wien
Registration: If you would like to join for this event, please email to: vsc-seminar@list.tuwien.ac.at
Abstract:
MPI is a large and complex, but well-structured interface for message-passing based parallel programming for high-performance systems. This tutorial aims to provide an understanding of basic concepts of the interface, showing how concepts and (advanced) features of the interface can be put to work effectively in applications and lead to more read- and maintainable, better performing and more robust code. More advanced features that will be covered include non-blocking collectives, neighborhood collectives, one-sided communication, derived datatypes and process topologies. The tutorial will provide room for discussion and interactions.