Visit additional Tabor Communication Publications
November 10, 2006
Two weeks ago at the ECMWF (European Centre for Medium-Range Weather Forecasts) Workshop in Reading, over 120 meteorology experts, computer practitioners and vendor representatives spent four days exchanging experiences about the latest results in meteorology and the computer infrastructure which goes along with it. This relatively small and friendly workshop provided a forum for the créme-de-la-créme of HPC users. What followed was a tour de force in meteorological and computing techniques by active practitioners striving to maximise the latest HPC technology and to refine their weather and climate forecasting models. They presented today's practical reality with teraflops systems, as well as their aspiration and vision for petaflops computing in the next five to ten years.
The Workshop included some 35 presentations and a discussion panel. Most of these presentations were by experts from major meteorological centres, from the U.S., Canada, Brazil, Europe, Japan, Korea and China. The rest were from HPC vendors -- Cray, Fujitsu, IBM, Linux Networx, NEC, SGI and DataDirect Networks. Thursday afternoon was devoted to a brain storming debate, hoping to identify solutions to the many pressing needs of this increasingly important field of science.
With almost every presentation meriting a treatment of its own, the selection of material in this article is biased towards meteorology applications, and although somewhat arbitrary, hopefully captures the essence of what was presented as well as highlights some news items.
In recent years, meteorology evolved from its esoteric weather prediction role to become a high profile e-business with enormous commercial clout. With climate change manifesting itself in extreme weather patterns, be it droughts, rain floods, or, more destructive hurricanes, the economic stakes are high. The field of meteorology can marshal large budgets -- $25 billion, 2001-2007 -- needed for data collection, assimilation and the purchase of large-scale computer systems for numerical modelling. So naturally computer vendors are keen to participate in the deliberations and offer previews of their future products.
In weather and climate numerical modelling, the debate on price/performance of commodity systems verses special-purpose systems should be irrelevant. When a hurricane is heading for land, capability computing at a level required to deliver advance predictions in time to implement protection procedures is the only measure worth considering. The imperative is for the ocean model to run on a fine mesh, in hundreds of metres say, not in tens of kilometres, and deliver results on time. The work presented by Keiko Takahashi using the Earth Simulator demonstrated the promising insights one can get from using high resolution models. Their work uses a non-hydrostatic coupled atmosphere-ocean simulation code with 100m to 2km resolution to predict local heavy rain days or weeks in advance, and 2km to 10km resolution for global seasonal predictions. The simulations promise to predict reality very accurately and become a toolkit for survival.
The first presentation, "development of the ECMWF forecasting system" was by Adrian Simmons from ECMWF. He started with the famous picture (circa 1911) from the book of Lewis F. Richardson (1922) of a myriad of humans calculating together (as a computer) to perform numerical weather prediction with processes. Richardson's fantasy had to wait until Charney, Fjortoft and Von Neumann used the ENIAC electronic computer in 1950 to take the first step by performing the numerical integration of the barotropic vorticity equation.
ECMWF became operational in 1979 with the mission to provide medium weather forecasting for member states from Europe. The advances in numerical forecasts are umbilically connected with computer developments. Increases in computer power, has enabled increases in vertical and horizontal grid resolution, more sophisticated analysis of observations, more realistic representations of atmospheric physics and land surfaces and the coupling with ocean wave and circulation models. Delivery of forecasts is more timely and comprehensive including probabilistic ones, based on ensemble methods. There are also new monthly and seasonal predictions as well as reanalysis of multi-decadal observations for understanding long-term weather trends.
In 1979 ECMWF operated a 50 megaflop Cray-1. In 1990 it had a 1 gigaflop Cray Y-MP-8. Today it operates an IBM P5-575 with 4480 CPUs delivering 4 teraflops of sustained performance across their codes. In 1979, their forecast used 200km horizontal grid and achieved 70 percent accuracy for a 3-day-in-advance prediction and around 37 percent with a 7-day prediction. In 2006, the 3-day prediction improved to 96 percent and the 7-day prediction improved to 70 percent.
In 1979, 24-hour data assimilation took 20 percent of computer usage and the rest was used for the deterministic forecast. Now 22 percent of computer usage is used for data assimilation, 17 percent for the deterministic forecast, and 61 percent for ensemble probabilistic forecasts. Since 1979 and the introduction of satellite data capture, the cost of data assimilation relative to single forecast increased more than ten-fold.
The numerical weather prediction (NWP) field has benefited from better models, higher resolution, improved representation of physical processes (such as radiation) and more comprehensive dynamical equations, incorporating chemistry and a sea-ice component. It uses improved methods in determining initial conditions for deterministic and ensemble forecasts, increasing the utilization of satellite data, with larger window for data assimilation. It takes a more unified approach to prediction, bringing more, ocean into the earlier part of the forecast range and incorporating extra aspects of air quality into the core forecasting activity.
The technical challenges facing meteorologists are how to effectively utilize the increasing numbers of computer cores. Larger problem sizes help, but the number of model points per core decreases as resolution increases for given execution time -- a problem also for ensemble forecasts due to memory constraints. The core speed and memory gap increases communication and load-imbalance overheads.
Load balancing also becomes more challenging as models include a wider range of processes. Assimilation of observational data poses substantial additional challenges. It involves repeated mapping between observation and model space, iteratively adjusts the model at a lower resolution than primary forecast, and has higher I/O demands.
Other challenges include ensuring continued effectiveness of algorithms that today balance accuracy and efficiency (at the expense of a lower ratio of sustained to peak flops), with the scope for refinements in design and implementation and perhaps for more radical change, but nevertheless are subject to limits imposed by physical laws and the nature of remotely sensed observations.
There is also the desire to ensure continued effectiveness of long-lived codes in which there has been major investment. For example, the joint ECMWF / Météo-France code originated in 1987, has run operationally on Cray C90 vector shared memory, Fujitsu VPP vector distributed memory, IBM scalar SMP cluster and will run operationally on NEC SX8R vector SMP cluster.
Another talk, given by Anthony Hollingsworth from ECMWF, described the Global Earth-system Monitoring using Space and in-situ data (GEMS). This project was started in 2005 and its overall objectives are to: 1) Exploit the huge investments in satellite data; 2) Extend NWP modelling and data assimilation capabilities to atmospheric composition on global and regional scales; and 3) Provide a new range of services for Europe, with global and regional deliverables.
The motivation for GEMS is treaty assessment and validation. Conventions such as Kyoto, Montreal, LRTAP and IPCC need best estimates of sources, sinks and transports of atmospheric constituents. GEMS should also enable the provision of better operational services to the community from improved forecasts. For example, in the 2003 summer heat wave, there were 18 thousand deaths in France and at least 33 thousand in Western Europe. Many would have been avoidable with advance warning.
On the macro level, environmental concerns have triggered $25 billion for new satellite missions in 2001-2008. GEMS plans to synthesise all available satellite and in-situ data into accurate 'status assessments', and will meet many needs of the GCOS implementation plan.
The global GEMS system is organised in six projects. Data input (assimilation and evaluation), greenhouse gases, reactive gases, aerosols (sea salt and desert dust), regional air quality and validation. The aim is to provide initial and boundary conditions for operational regional air-quality and 'chemical weather' forecast systems and provide improved monitoring and forecast services for the health sector on UV exposure and skin cancer, heat stress and drought, acute pollution events, respiratory and cardiovascular disease, and in the future, vector borne and zoonotic disease (e.g. malaria experience). It will also provide regional estimation of sources and sinks of carbon dioxide, ozone, aerosol and so on.
Global forecast of pollutant concentration for the media, the public or city specific, are embodied in the scheme. The production of regional forecasts of chemical species and air quality indices based on an ensemble of air quality models on the European scale are already in their research phase. For example, surface ozone daily maxima are issued from various research sites in France and Germany as illustrated by results from 20th October 2006.
ECMWF plans to have an operational global monitoring forecast system for atmospheric composition, combining all remotely sensed and in-situ data to create three-dimensional global distributions [50km (H), 1km (V), 6-hours] by 2009. The key atmospheric trace constituents of this system are: 1) Greenhouse gases -- initially CO2, and progressively adding CH4, N2O, plus SF6 and Radon to check advection accuracy; 2) Reactive gases -- initially including O3, NO2, SO2, CO, HCHO, and gradually widening the suite of species; and 3) aerosols -- initially a 15-parameter representation, later around 30.
GEMS will provide a retrospective analysis of all accessible in-situ and remotely sensed data on atmospheric dynamics and composition for the ENVISAT-EOS era (1999-2007). It will also provide monthly seasonal maps of the sources, sinks and inter-continental transports, of CO2, O3 and many other trace gases and aerosols, based on in-situ and satellite data.
Bill Kramer from NERSC/LBNL, gave his presentation: "NERSC experience: Implementation of a facility wide global file system". He briefly touched on favourites such as the widening gap between application performance and peak performance of high-end computing systems; the recent emergence of large, multidisciplinary computational science teams in the DoE research community; the flood of scientific data from both simulations and experiments; and the convergence of computational simulation with experimental data collection and analysis in complex workflows.
He then went on to talk about GUPFS (Global, Unified, Parallel File System). This is a multi-year projec, started four years ago, to deploy a centre-wide shared file system at NERSC. The aim is to make advanced scientific research using NERSC systems more efficient and productive, simplifying end-user data management by providing a shared disk file system in the NERSC production environment. GUPFS is global and unified. The file system is shared by major NERSC systems using consolidated storage and providing a unified name space. It automatically manages the sharing of user files between systems without replication. NERSC has a version of GUPFS in production for over a year and it works. The plan is to integrate it with HPSS and Grid. Their target mission is to have a parallel file system providing performance that is scalable, as the number of clients and storage devices increase.
"We have a path forward that allows all computer architectures to participate fully," said Kramer. "There is already a huge benefit to a number of users. Two years from now, we expect to report all systems and users are using the global file system, many exclusively".
In the course of this Workshop, many other presentations dealt with solving substantive problems by coupling weather and ocean models. Community frameworks for building coupled Earth System Models have been an area of intense research and development over the last few years. The GDFL Flexible Modelling System (FMS) has been in active use for about five years. Two broad-based efforts to develop frameworks across the community are now approaching maturity levels that allows for actual deployment: The Earth System Modelling Framework (ESMF) in the U.S. and the European Network for Earth System (ENES) Modelling in Europe.
At the other end of the spectrum, forecasts are used to validate the U.S. inter-organizational modelling initiative known as the Weather Research and Forecast (WRF) model. WRF has a three-pronged objective of developing 1) the next generation meso-scale Numerical Weather Prediction (NWP) modelling system for research and operations; 2) a common modelling infrastructure that facilitates operational NWP collaboration, scientific interoperability, accelerates the transfer of new science from research into operations; and 3) a repeatable process that continuously infuses innovations and capabilities into the community meso-scale NWP modelling system.
As principal partners of this U.S. national effort, the Air Force Weather Agency (AFWA) and Fleet Numerical Meteorology and Oceanography Centre (FNMOC) have been able to leverage a vast array of resources only available to Department of Defense (DoD) entities, in particular, the computational resources (340 teraflops) made available through the DoD's High Performance Computing Modernization Program (HPCMP), whose objective it is to facilitate the rapid application of advance technology into superior war-fighting capabilities. Douglass Post, chief scientist for HPCMP, gave an interesting progress report on the challenges for delivering such complex computational tools for mission-critical activities. This is a core element of the U.S. doctrine for acquiring full-spectrum dominance of military capabilities, enabling it to impose unilateral solutions at will. Douglass' message was that the DoD needs to generate an infrastructure with more engineering and less computer science experimentation if their computer-based experiments are to be taken seriously. They need code architects and rigorous verification and validation schemes, from the start of a project. For DoD applications run in production, teraflops and even petaflops computing power is essential, as timeliness is critical.
The vendors described their latest products and near future research and development activities, moving towards petaflops capability computer systems. Power consumption, reliability of 100,000-plus cores and efficient software infrastructures for ease of use, are major challenges for all vendors straddling the commodity chip roadmap.
To revert to meteorology and slightly change Lewis F Richardson's preface of his 1922 book: "Perhaps ... in the near (dim) future it will be possible to advance the computations faster than the weather advances and at a cost less than the saving to mankind due to information gained."
I wonder which brave meteorologist would visit Richardson's grave in 2011, salute and say: "We played with your fantasy. We imagined a large hall, like a theatre... A myriad [human] computers went to work upon the weather of the part of the map where each one of us sits... Numerous little night signs display the instantaneous values so that neighbouring computers can read them... and after a century, I can report: Mission accomplished, Sir." The intriguing thought occurred to me: Will Professor Richardson, understand the language the message is delivered in?
Finally, enjoy SC06 in Tampa. Sadly I will not be joining you.
Copyright (c) Christopher Lazou, HiPerCom Consultants, Ltd., UK. November 2006. Brands and names are the property of their respective owners.
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
May 10, 2013 |
Program provides cash awards up to $10,000 for the best open-source end-user applications deployed on 100G network.
May 09, 2013 |
The Japanese government has revealed its plans to best its previous K Computer efforts with what they hope will be the first exascale system...
May 08, 2013 |
For engineers looking to leverage high-performance computing, the accessibility of a cloud-based approach is a powerful draw, but there are costs that may not be readily apparent.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.