The Earth Simulator - A 40 Tflop/s supercomputing Grand ChallengeLondon 19 June 2000 In an interview by Christopher Lazou, UK Editor of Primeur, Dr. Keiji Tani, one of the principal scientists involved with the Earth Simulator said: "The Earth Simulator is a parallel vector processor system designed to create a virtual earth for scientific research in global warming; earth disasters caused by phenomena such as el Nino as well as seismic and volcanic activities. It is an inter-discipline project, at the centre of Japanese research and with some of the Japanese most prestigious organisations involved. International collaboration is actively sought for the scientific part of the Earth Simulator project which is expected to come on stream in March 2002". Read the full interview.
CL: (Christopher Lazou): What is the core vision of the Earth Simulator Project?
KT: (Keiji Tani): The Earth Simulator project has at its core the vision of attempting to predict Global change. This is the most pressing issue of today, for achieving sustained development of humanity across the globe and in protecting life from natural calamities.
CL: What does it bring together which is new and different from current national research?
KT: The Earth Simulator integrates three elements, a supercomputer hardware facility with Tflop/s performance, basic research in the development of physical models for Global change prediction, observations via satellite, (TRMM) and oceanographic research vessels (MIRAI), into one coherent project.
The STA(Science and Technology Agency of Japan) is promoting studies on the global change prediction with an three-in-one research system. As part of the system, we are developing the Earth Simulator. Please note that the Earth Simulator is one of the elements.
CL: I understand this is a collaborative project, so who are you collaborating with?
KT: From its inception, the Earth Simulator envisaged strong international collaboration for developing the physical climate models and collecting the observed data. We have just began discussions on the software with many experts and expect strong international input. This close scientific involvement and ownership of results is essential in order to achieve international consensus for political action. If it remained within Japan the international consensus will not be forthcoming.
CL: Why is consensus so important?
KT: As one knows weather prediction and climate changes, apart from impinging on social and economic fields it also impinges on national security. Thus international consensus is essential for the success of the project, with its lofty aim of influencing government strategy and their attitude to international decisions which would have to be taken in order to manage Global change.
CL: Where does the Earth Simulator fit in the research structures of Japan?
KT: The Earth Simulator is an inter-discipline project, at the centre of Japanese research and with some of the most prestigious organisations involved. These include, the National Space Agency of Japan, (NASDA), Japanese Marine Science and Technology Centre, (JAMSTEC), Japanese Atomic Energy Research Institute,
(JAERI), the Research organisation for Information Science and Technology (RIST) and the Institute of Physical and Electrical Research (RIKEN). Many other institutes and centres for software development, as well as, the Japanese Meteorological Office, and researchers in universities are also involved. The operational aspects are delegated to JAMSTEC.
CL: What is the current status of the project?
KT: The buildings are well underway, so the hardware facility is likely to be ready by early spring of year 2001. We expect the software will take a bit longer, but should be running some preliminary models soon after. The key priority for us, is to put the infrastructure in place, to enable the scientists to run their models and develop new more complex ones.
CL: What are the supercomputer hardware you are putting in place?
KT: We evaluated both the scalar MPP and vector parallel systems and chose the most appropriate for our application. The hardware for the first version of the Earth Simulator are fixed and based on NEC chip technologies. The architecture is a MIMD type distributed memory parallel system, consisting of computing nodes with shared memory vector type multiprocessors. It consists of 5,120 processors, in 640 nodes with 8 processors per node. The nodes are connected using a single-stage crossbar network. The total main memory is 10 TeraBytes, with 16 GBytes of shared memory in each node. Each processor has a peak performance of 8 Gflop/s which translates to a total of 40.96 Teraflop/s peak performance for the whole system.
CL: How did you arrive at the system performance size?
KT: The world is faced with climate change characterised as global warming. More specifically we have meteorological disasters such as heavy rain (monsoon), droughts, air pollution, acid rain, ozone depletion and so on, and in addition we have earthquakes and volcanism. These factors in the extreme threaten the sustained
development of human beings and in certain circumstances they can even change the course of evolution. We therefore, identified the following targets:
1. Applications to atmospheric and oceanographic sciences; using high resolution global models for prediction of global warming etc.; using high resolution regional models for prediction of El Nino events and Asian Monsoon etc.; high resolution local models for prediction of weather disasters such as typhoons, localised torrential downpours, oil spills, down bursts etc.
2. Applications to Solid Earth sciences; a global dynamic model to describe the entire solid earth as a system; a regional model to describe the crust/mantle activity in the Japanese Archipelago region; a seismic wave tomography; a simulation of earthquake generation processes; and a simulation of material transport in strata.
CL: Is that how you arrived at the requirements for the Earth Simulator?
KT: Yes; One looked at the present requirements for existing models and scaled the
requirements when using more refined models, for example, reducing the horizontal
mesh distance from the present 50-100 Km to 5-10 Km. Proceeding in this way one
deduces that the memory requirement is around 3.84 -7.68 TeraBytes. In a similar way
one calculates that the sustained effective performance of the Earth Simulator must be
more than 5 Tflop/s. For the CPU we assumed 12.5% efficiency, a conservative
estimate, but nevertheless, both CPU and main memory have to be at least one
thousand times larger than those of present computers used for atmospheric
circulation models. So you can see how we arrived at 40 Teraflop/s and 10 TeraBytes
of memory for the Earth Simulator.
CL: In the USA a lot of effort has been put in the ASCI programme, part of the
Nuclear military requirement, which paved the way for the development of High
Performance Computers which deliver Teraflop/s performance. Is the Earth Simulator
the peaceful alternative engine to ASCI?
KT: Yes, as a matter of national policy, Japan does not have, or, plans to develop
nuclear weapons. "Our activity with the Earth Simulator is a totally peaceful one and
is mainly addressing a problem of global change for the 21st century."
CL: Having access to cutting-edge computing, namely the largest vector parallel
system with Teraflop/s, how much more performance would you need in the kind of
application you are involved?
KT: The Earth Simulator envisages a 10 Km grid; scientists ideally would like to
refine their models down to 1 Km. This requires a thousand times more performance
than planned for the Earth Simulator. This translates to a requirement of at least 40
Petaflop/s peak or more than 5 Petaflop/s of sustained performance and 100 Petabytes
of main memory.
CL: Where do you think this technology is leading us?
KT: The Earth Simulator's hardware use chips with 0.15 micron etchings. In five to
seven years it is expected that this will reduce to 0.05 microns enabling chip
fabrication with 1 Giga of transistors. This type of density may be possible, but it is
difficult to predict where it will lead, since at this density many other engineering
problems are spawned and need to be addressed. The scientific market is too small so
one assumes the consumer market will have to drive these developments.
CL: What about languages and system software requirements?
KT: At present we plan to use an extended version of High Performance Fortran,
JAHPF2. This should satisfy 80% of calculation of results. We also plan to provide
special system software to optimise the use of the Message Passing Interface, MPI2,
and to assist with parallelism, tuning, monitoring and debugging.
CL: If money and technology were no barrier what would be your ideal system for
running your present model and perhaps allowing for new physics to be incorporated
in the model? E.g. biosphere multi-scale and multi-physics simulations?
KT: As explained above, even with current models scientists have requirements for
Petaflop/s. For new more complex models incorporating new physics, including
biosphere life variables, the computing requirements will snowball and should keep
computer engineers and scientists busy for many decades and possibly centuries.
CL: Finally, evidence is mounting which shows that man made activities are
stimulating global warming with life threatening effects. This can be seen in the
stressful effects on polar bears in the Hudson Bay, the extinction of the golden toad in
Costa Rica, to the destruction of the coral reef in the Indian Ocean. Are you optimistic
that results from the Earth Simulator project which may shed light on the underlying
changes which are taking place would be available in time to save the world from
destruction?
KT: This is difficult to predict. One has to make a beginning and search for
fundamental answers. It also requires educating the people to appreciate the dangers
and the political will to take remedial actions on a Global level. There are many
difficulties to be overcome, this is why we in Japan recognise that the Earth Simulator
has to have a strong international dimension to have any chance for success.
CL: I think we explored a number of issues. Thank you for your answers which I am
sure readers would find very interesting.
For further details contact:
Dr. Eng., Keiji Tani,
Copyright: Christopher Lazou, Managing Director, HiPerCom Consultants Ltd., UK, and Author of
"Supercomputers and their Use", OUP, 1989.
e-mail: Chris@lazou.demon.co.uk
Chris Lazou [News on Advanced IT] [Calendar] [Analysis] [IT in Medicine] |
 |