Cray User Group Focuses on Scalability for Weather Forecasting

Print Friendly, PDF & Email

ungaroGlobal efforts to bring about crucial improvements in supercomputing efficiency and energy usage were placed center stage this week as the European Centre for Medium-Range Weather Forecasts (ECMWF) welcomed users and vendors from around the world to London for the Cray User Group 2016 conference.

The Cray User Group is a vibrant, user-managed community and its annual conference provides a unique forum for discussing and sharing solutions to some of the most challenging problems that are at the intersection of advanced technology, supercomputing and big data analytics,” said Cray president and CEO Peter Ungaro. “We appreciate the support of ECMWF in hosting this year’s excellent conference, and we look forward to working closely with ECMWF and our customers around the globe to meet the scalability challenges of tomorrow.”

ECMWF, which runs one of the largest supercomputer facilities in Europe, is a leading proponent of ‘scalability’, a series of initiatives designed to improve supercomputer efficiency and energy usage. Innovative ways to allow high- performing computers to work more efficiently will enable greater amounts of data to be processed. This in turn will allow crucial weather forecast and Earth monitoring improvements to continue whilst sustainably managing the amount of energy needed. As a result, ECMWF’s numerical weather predictions will continue to support Europe’s meteorological services as they protect their citizens from the effects of severe weather.

Scalability was the key topic at this year’s CUG, which brought together users of Cray supercomputers from all around the world. The event included tutorials and technical sessions on how best to use high-performance computing systems, and to discuss the importance of improving scalability in the area of numerical weather prediction.

ECMWF’s first operational forecast in 1979, at 210 km global resolution, took 5 hours to run on a single-processor Cray-1A. Today, two Cray XC30 systems, each with more than 84,000 cores, provide more than 4 million times the sustained performance of that first Cray. This allows a much larger and more advanced forecasting system featuring a high-resolution global model resolution of 9 km and a 50-member ensemble with a model resolution of 18 km.

chart

Further improvements in forecasting will require supercomputer energy consumption to increase in order to manage the more complex forecasting systems of the future. Current projections show that energy consumption at ECMWF would have to increase from around four megawatts today to around 50 megawatts within ten years. Scalability is intended to reduce this and keep supercomputers energy efficient as they contend with more data.

Improvements in predicting the weather from days to months and in understanding our climate call for ever more complex calculations,” said ECMWF Director-General Florence Rabier. “Supercomputer technology has played a key role in the great advances we’ve made over the past four decades. “But this same technology could ironically impair further advances unless we start tackling inefficiencies in data processing and energy use.

workshop“The high interest in the presentations and discussions during the conference is encouraging and confirms the importance of the topic of scalability. The conference has been an excellent opportunity to raise the profile of ECMWF’s Scalability Program and has opened up the prospect of new collaborations in this area. If we can get this right, we won’t just be able to predict the weather more accurately further in advance and improve the resilience of our economies, we’ll also do it at a sustainable cost.”

ECMWF is also putting scalability at the top of the agenda for the 17th Workshop on High-Performance Computing in Meteorology, which will take place Oct. 24-28 in Reading. The event will bring together users from Member States and around the world to discuss achievements in high-performance computing over the past two years and requirements for computing power into the future.

Sign up for our insideHPC Newsletter