Deputy Director of Computing, European Centre for Medium-Range Weather Forrcasts (ECMWF)
One of the oldest sciences in existence is that of predicting the weather. In modern days, it’s become among the most advanced supercomputing disciplines, with complicated forecasting models requiring large amounts of compute muscle. One of the most advanced comes out of the ECMWF, who supports 34 European nations with high resolution forecasts and other climate related research. We caught up with Isabell Weger, Deputy Director of Computing at the Centre, to see what’s up in 2014.
HPCwire: Isabella, you head up what is considered one of the most advanced HPC weather centers on the planet, with two IBM Power7 775 clusters installations. Can you tell us about some of the compute hungry initiative and models that you use that require this amount of power?
Isabella Weger: ECMWF’s HPC facility, one of the largest in Europe, is at the core of our activities. As the world’s leading centre for global medium-range weather prediction the delivery of a consistent and reliable 24/7 service is key – the dual-cluster installation provides resiliency for the forecast production and allows efficient use of resources for research and operations. Substantial HPC resources are required to run operational forecasts to a strict production schedule and to perform research experiments to advance weather prediction techniques and products.
ECMWF’s Integrated Forecasting System (IFS) uses sophisticated numerical models to simulate the evolution of the complex processes in the atmosphere and ocean. ECMWF produces global 15-day forecasts twice per day. Starting point is the information about the current state of the weather worldwide – more than 40 million observations per day (from weather stations, ships, aircraft, balloons, satellites etc) are analyzed and fed into a physical model of the atmosphere. With a 16km horizontal resolution and 137 levels in the vertical, the model forecasts wind, temperature and humidity at nearly 300 million grid points throughout the atmosphere at each time step. In addition, an “ensemble prediction”, a set of 51 forecasts from perturbed initial conditions is produced to provide an estimate of forecast uncertainty.
ECMWF also produces monthly and seasonal predictions and “reanalysis”, combining past meteorological observations into comprehensive atmospheric data sets for climate monitoring and research.
HPCwire: Weather modeling and simulation has always been one of the primary uses of HPC. I’m not sure that people necessarily appreciate the amount of evolution that goes on with the discipline as technology and data collecting methods advance. Can you talk about the evolution and changes you see on the horizon related to the tech advances you’ve witnessed?
Isabella Weger: Advances in HPC technology and performance allow a more comprehensive earth-system model with the atmospheric model being coupled to ocean, wave, sea ice and land-surface models; inclusion of atmospheric composition (e.g. aerosols and greenhouse gases) and prediction of air quality; and better early warnings for severe weather events, such as windstorms, tropical cyclones, floods, heat waves etc.
Numerical weather prediction has always required using leading-edge technologies. The main drivers for compute performance are finer model grid resolutions, assimilation of more observations and better representation of physical processes in the atmosphere.
ECMWF’s first operational forecast in 1979 was run at 210 km global resolution on a single-processor Cray-1A. The current model resolution is 16km, and the next HPCF will allow us to introduce a 10km grid resolution (and more advanced science) in 2015.
Clearly, HPC technology developments are influencing the directions for our research will take. ECMWF’s current HPCF has in total nearly 50,000 cores; its successor will have about 160,000 cores. Advanced ensemble systems are one way to utilize more cores. However, increasing core counts will enable – and require – much increased levels of parallelism in our data assimilation system and forecast model. Exploiting parallelism on all architectural levels and improving the scalability of all codes will be both vital and challenging for NWP.
HPCwire: As you look down the road at 2014 and beyond, what are the trends developing in high performance computing that you see as important now and into the future for your discipline?
Isabella Weger: Energy efficient HPC and flops/Watt are driving developments towards massive core counts, accelerators and heterogeneous architectures. NWP applications must exploit parallelism on all levels to improve their scalability. ECMWF’s IFS is already a highly parallel software using an MPI/OpenMP hybrid parallelization scheme and is very efficient on current architectures – experiments at very high 5km resolution have been run across 200k+ cores.
The future use of accelerators and many-core processors is both promising and challenging. ECMWF’s IFS reflects leading scientific developments in numerical weather prediction over more than 30 years in an application with 1.5 million lines of code; portability across the prevailing HPC architectures has always ensured healthy competition in procurements. With advanced architectures, this becomes even more a challenge, and high-level approaches such as libraries, OpenACC and compiler/software tools will be important.
ECMWF is engaging in an ambitious scalability project for the future development of NWP applications. New, and potentially disruptive, scientific and algorithmic approaches and programming techniques will be required. Scalability in IFS is limited by global communications, alternative data structures will have to be explored with a focus on localization of data, both in the spatial and temporal domains.
On a completely different front, “big data” is also relevant for weather forecasting, as it requires and generates a massive amount of data. ECMWF has built a unique archive that holds the largest collection of meteorological data worldwide, about 50 PB and growing by 70 TB/day.
HPCwire: On a personal note, can you talk about your personal life? Your family, background, any hobbies?
Isabella Weger: I am from Austria, but came to live in England together with my husband, when I was offered the position as Head of Computer Division at ECMWF nine years ago.
I enjoy spending time with friends and travelling with my husband. I got interested in English garden design and landscaping, and I like to be outdoors – unless the English weather really lives up to its reputation.
HPCwire: One last question – is there anything about yourself that you can share that you think your colleagues would be surprised to learn?
Isabella Weger: Yes – I am not on Facebook, and I don’t tweet.