Tuesday 30th May 2017

Webinar : Approaches to tackling I/O issues at large scale : Wednesday 14th June 1500 BST

Adrian Jackson and Nick Brown, EPCC look at using MPI-IO and doing inline analysis.

Adrian: Improving COSA at scale: Parallel I/O and load balancing.
This presentation will outline the work we have undertaken on the harmonic balance CFD code, COSA, to improve both performance on large core counts and improve the usability of the code. The original parallel I/O functionality in COSA, whilst able to write large amounts of data in parallel, was sub-optimal in its use of MPI-I/O. We have optimised this, reducing the I/O time by up to 70% on large core counts.

Furthermore, the domain decomposition used to split work across cores in the parallelisation meant users had to work hard to ensure the simulation input was load balanced. We have added functionality to automatically load balance simulations, meaning users have to undertake much less work preparing simulations, speeding up the whole science workflow for this application, and ensuring resources are efficiently used when running simulations.

Nick: In-situ data analytics for atmospheric modelling
The Met Office NERC Cloud model (MONC) has been developed by EPCC and the Met Office to enable weather and climate communities to model clouds and turbulent flows at scale on thousands of cores. The computation itself generates raw prognostic 3D fields, however the scientists are far more interested in higher level, diagnostic, information such as how the average temperature at each level in the system changes over time. As such, MONC also has to support data analytics but based upon the very significant (many GBs) size of the raw fields involved we do not want to pay the cost of physically writing raw data to disk or pausing the computation whilst analytics is performed. Bit reproducibility, performance & scalability, and the ability to easily configure and extend the analytics meant that we decided to develop our own approach.

In this talk I will discuss the motivation in more detail, describe our analytics architecture, some of the specific challenges we faced and solutions to these, and discuss performance and scalability results of a standard MONC test case on ARCHER running over 32768 computational and 3277 analytics cores.

More information and connection details...

Women in HPC at ISC 2017, Frankfurt, Germany

Full information is available at: Women in HPC at ISC

ISC is now only 3 weeks away. When you plan your schedules, mark your calendar for the WHPC workshop and BoF.

BoF: Practical Steps to Diversifying the HPC Workforce
Wednesday 2rd June 2017, 1.45-2.45pm
Substanz 1+2, Frankfurt Messe

Following the success of our BoF at SC16, we will have a practical audience-focused BoF at ISC17 discussing unconscious bias, stereotype threat, or imposter syndrome. The BoF will discuss the real effects of these topics on the workplace, providing the audience with an introduction to each theme, how these themes may affect you, and how they impact employers, employees, advisors, managers or your peers. This discussion will also be of benefit when considering improving diversity for underrepresented groups other than women.

Workshop: Diversifying the HPC Community
Thursday 22nd June 2017, 9 am - 1 pm
Marriott, Frankfurt

Our 6th international WHPC workshop will continue with the theme of educating and supporting diversity in the workplace.
This workshop will focus on two topics:

  • Skills to thrive in an HPC career.
    Here all participants get hints and tips on public speaking, effective workplace communications, taking your career to the next step, and speed networking practice.
  • Improving diversity in the workplace.
    These sessions are targeted for for anyone in the workplace, with specific tools for leaders and managers on how to successfully address discrimination in the workplace, the benefits of professional networks and mentoring, and how to identify the roadblocks facing those in underrepresented groups.

Everyone is welcome to all of WHPC's events irrespective of gender, career stage, or background. We have a goal to become more diverse ourselves and particularly welcome male attendees. It takes all of us working together to build better performing and more productive teams.

Upcoming Training Opportunities

  • Scientific Computing 31st May, 7th, 14th & 21st June 2017 Edinburgh and Online
  • Approaches to tackling I/O issues at large scale 14th June 2017 Online
  • Shared-Memory Programming with OpenMP 12th -13th June 2017 Imperial College London
  • Scientific Programming with Python 19th - 20th June 2017 EPCC Edinburgh
  • ARCHER Champions 26th - 27th June 2017 STFC, Daresbury
  • ARCHER Summer School 2017
    • Hands-on Introduction to HPC 10th - 11th July 2017 Edinburgh
    • Message-passing Programming with MPI 12th - 14th July 2017 Edinburgh
  • Shared-Memory Programming with OpenMP 1st - 2nd Aug 2017 Oxford
Full details and registration...