We recently joined the European Technology Platform for High Performance Computing (ETP4HPC) and attended our first European HPC Summit Week 2017 in Barcelona, the European Extreme Data & Computing Initiative, EXDCI. The programme for the week contained workshops from various European HPC initiatives in order to find synergies among all parties of the HPC ecosystem.
The week started with the EXDCI workshops along with Partnerships for Advanced Computing in Europe, PRACEdays17, finishing with ETP4HPC and EUDAT workshops. The most striking subject that resonated with me (due to being a data center provider), was that of improving supercomputing efficiency. It was also very much a common theme through a number of the presentations, seminars and panel discussions. This is a wide-ranging subject that applies to all aspects of the supercomputer and the personnel involved within this complex ecosystem.
With Exascale computing (systems capable of at least one exaFLOPS, or a billion billion calculations per second) targeted for 2020 – 2023, the HPC Summit EoCoE workshop sought to understand how best to use these efficiently. Some of the many applications these machines will be used for, are maximising the efficiency of solar arrays, hydro-electricity and geothermal plants along with wind farms.
There were several subjects that were raised, highlighting the inefficient design, use, workflow, application design and application management across European HPC deployments. There is friction between the hardware builders and the application users who are not maximising the most efficient use of the HPC system. The challenge is deploying an HPC platform efficiently, that can support the different disciplines from physics, chemistry, biochemistry, weather and renewable applications.
Professor, Marc Snir of Computer Science at the University of Illinois, explained how energy consumption is being reduced by increasing denser packaging, ensuring the compute is as close as possible to the memory, utilising specialised compute engines and targeting the reduction of chip energy by two times. Marc went on to suggest that further efficiencies can be made with a return to liquid nitrogen as chip manufacturing becomes denser.
Professor Dr. Paul Gibbon’s, Head of the Division "Computational Science" at the Jülich Supercomputing Center, presentation explained the part that HPC plays in energy systems modelling for solar arrays. Paul described how the Weather Research and Forecasting (WRF) community model has provided an appropriate numerical weather prediction framework for solar power day-ahead forecasting. The challenge includes the solar multiple, number of storage hours, system efficiencies, location and local energy markets.
Paul also described how Germany is reducing its reliance on fossil fuel and how it has achieved this through renewable sources at certain times during the year. This is possible through the investment in solar arrays and wind farms. As with the solar array predicative applications and day-ahead forecasting, this can also be achieved with the wind farms, by ensuring the height, pitch angle and placement of the turbine blades are at the optimal position to harness the wind. However, as these energy sources aren’t constant, a data center housing an HPC platform cannot rely on this source of renewable power and will have to revert to traditional, fossil fuel power plants, as the Jülich Supercomputing Center does from the RWE Weisweiler lignite power station, which produces 18 megatons of CO2/year, with a planned lifetime to circa 2045.
With Exascale computing targeted for 2020 – 2023, “the billion-core machine”, will further exasperate these energy use challenges and driving efficiency throughout the supercomputer ecosystem will be critical. A great deal of effort is going into this at the hardware and application level. It occurs to me that future Exascale machines should have a power foundation supported by low cost, constant renewable energy to allow the scientists and hardware providers to build on this, with the most efficient compute applications and deployment. HPC use is set to grow across both research and the industrial community. We are in danger of consuming more fossil fuels, through the explosion of this technology and undoing the improving efficiencies the renewable energy applications provide for these renewable plants across Europe.