Slashdot Mirror

Okay... This is what it's supposed to emulate.

This thing has more than nine hundred thousand processor chips and two petabytes of memory. Current x64 chips are limited to 256 TB (wikipedia) of physical address space; so these chips either [a] have larger than usual physical address space (I doubt), or [b] isn't a shared memory system.

So, dumbnuts, this isn't a shared memory system. Go read about the Cray XC40. Or even this document -- clearly showing it's a multi-node system with a fast interconnect. (It talks of each node running different OS images, so that means it isn't one shared OS image - which means it isn't shared memory).

Summary: What evidence do you have that the target system is shared memory? It looks to me like it's non-shared-memory (i.e., message passing); while with an extremely fast interconnect, I'm sure it's still slower than the CPU internal busses. The same is true with this Raspberry Pi - the interconnect (ordinary Ethernet) is still significantly slower than the ARM chip itself; and THAT environment is what's being emulated.

It doesn't really matter that other architectures could be faster - the GOAL is to replicate how the Cray XC supercomputers work - albeit at a fraction of the performance and price.

These simulations are forecasts. They check every forecast against observations, and have very good metrics on how good their forecasts are, and how much skill changes.
See for example how the European ECMWF does its forecasts:
http://www.ecmwf.int/en/foreca...

Every change to the operational model(s) can be and is checked out first against " will it improve the forecast". Similarly improvements in computing power: we simply run yesterdays forecast at higher resolution for example; we can then say "this new model is n% better, but takes 10x as long to calculate", and use that to decide whether its worth buying a faster computer.

On the climate timescale we have a challenge verifying the simulations, but on the weather timescale its straightforward, and done.

A bit of a pedantic point here, the European Center for Medium Range Forecasts weather prediction models are the most accurate of the current bunch. That said, I am not sure if this represents a sea change in how they ginned up the model or just the fact that they did it somewhat better than others (at present, anyway).

Further, I'm not sure that there is a substantive difference between European and American scientists in most scientific and engineering fields. There are roughly similar levels of Nobel Prizes, screwball research, entertaining products and outright failures on both continents. Cross your eyes a bit (or take off your glasses), and a Boeing plane looks just about like an Airbus.

In particular, European medical research is just as crappy as American attempts. But we are making progress, that's all that the scientific method gets you - moving in the 'right' direction. Slowly at times, but inexorably.

We're all bozos on this bus.

http://www.ecmwf.int/about/overview/
The population of the EU is some 500 million.

It appears that the computers that Europe was using for the "better forecast" were not as powerful as the old system being replaced. Upgrading because Europe's forecast better would be like taking a slow route to a holiday destination then buying a Porsche because your neighbours got there sooner when all you need is a new roadmap.

The problem you encounter is well known problem in meteorology. The two methods that are being used most are FE and spectral method. The former is simpler and more flexible and the later being more accurate since it keeps the derivations analytic. The problem with the spectral method is that you have to make your problem periodic, and you have to go from grid-point space to spectral space every timestep. In regard to stability, this is not a problem in many FE methods as long as you fulfill Courant-Friedrich-Levi criterion i.e. as long as your timestep is shorter than the time needed for the fastes wave to pass the distance between two grid points. When aplying FE method you have to make a choice of a scheme and depending on the scheme you can expect different distortions of your wave. Try to look in the book Numerical Methods for Solving Wave Equiations in Geophysical Fluid Dynamics by Dale Durran. Good on-line source is http://www.ecmwf.int/newsevents/training/lecture_n otes/LN_NM.html especially "Numerical methods" by R. W. Riddaway and M. Hortal

The Carnot efficiency of the possible energy captured from wind turbines from cooling ambient air in such a tower is around 2%. The thing is that Carnot efficiency is more relevant in non-renewable systems, because the supply of ambient temperature air is unlimited and by using a solar powered cooling system, there isn't energy expended in cooling the air.

The fact that the wind turbines can only convert 2% of the heat energy in the air into electricity isn't a consideration except for the physical construction costs of the tower. Once the tower is built, at least that portion of the system is little-to-no maintenance.

Utilizing a solar powered absorption heat pump increases the ROI of the solar panels by using the heat collected to move up to 20x the direct solar heat (the additional heat comes from the ambient air). The more air that has to move through the system, the better and the 2% Carnot efficiency isn't all that important.

Is http://www.hpss-collaboration.org/hpss/index.jsp something to try ?

The ECMWF uses this for their extremly large dataset. http://www.ecmwf.int/

I may not have made myself clear. It wasn't that Lorenz's model failed to predict the climate accurately, it was that his model failed to predict itself accurately. It wasn't a matter of choosing the wrong equations or data - it's a matter of self-referential, non-linear equations are inherently unpredictable. Bigger computers and bigger models don't get away from that problem.

I quote from the ERA-40 site: After 10 days of coupled integrations, the model drift begins to be significant.

It's exactly the same problem Lorenz had.

Both 24-hour and 50-year weather 'reports' use the same basic models and concepts

Actually, the ECMWF (European Center for Medium Range Forecasts) is trying seasonal forecasts for most world regions. The forecasts, as you can imagine, aren't very elaborate, and the accuracy is just a few points higher than folklore, but it's a starting point.

On the other hand, real weather prediction for more than, say, ten days is nearly impossible, since the "butterfly effect" grows to a size comparable to the modelled effects after this time.

One thing to the necessity of precise weather: I'm doing short-term forecasting of wind energy. There are regions in the world where in the area of one regional utility the penetration of wind energy is in the order of the minimum load encountered (like Denmark or north-western Germany). There, accurate forecasts can prevent your electrical grid from toppling over, or at least can prevent the utilities from having to buy expensive power on the spot market.