This page contains a Flash digital edition of a book.
HPC: ASTROPHYSICS AND COSMOLOGY


to having a giant supercomputer like Blue Waters, we couldn’t do these kinds of simulations in a reasonable timeframe.’ Since the last upgrade a few months ago, the Hubble Space Telescope has shown populations of extremely distant galaxies, as they were around 600m to 700m years after the big bang. ‘These are really early populations of galaxies, and it’s these galaxies that we’re trying to compare our results to,’ says O’Shea. ‘Hubble’s field of view is very tiny, and it will not see huge numbers of these galaxies – there are thousands of them, but they’re not very big, and they’re all very close together, because the universe was so small [that long ago],’ he says, adding that this is the reason that his simulation volumes do not need to be enormous. ‘We need to look at something 50m to 60m light-years per side in order to get all the galaxies we’d ever need.’ In terms of observational data, O’Shea is quick to point out that a galaxy such as our own Milky Way would not be visible at the distances involved here: ‘We’re always seeing the big weirdo, or the unusual cases. There are some galaxies that are forming stars at a really incredible rate, and if Hubble were 10bn light-years away, those are the galaxies that it would see.’ Nonetheless, he says, this raises the question of what an average galaxy looked like 500m years after the birth of the universe. A problem faced by O’Shea’s group and


‘If you make the cube big enough, then it’s statistically representative,’ says O’Shea, although he explains that bigger simulation volumes will not be the first use of the power offered by Blue Waters: ‘Going to a really powerful computer, we can include more physics,’ he explains, giving the example of adding hydrodynamic effects to the AMR- based simulation. Another area in which O’Shea hopes to see improvements is by including the effects of re-ionisation. ‘Stars, particularly massive stars, produce a lot of ionising radiation, particularly UV. This radiation ionises and heats up nearby gasses, and can actually stop star-formation around [the massive star]. What we typically do in these simulations is have a temperature knob for the universe that we turn up; it’s statistically representative of re-ionisation, but it’s not really accurate. There’s a lot of evidence that in the first billion years of the


www.scientific-computing.com


‘Dynamic range is the fundamental problem of cosmology… the universe is big, and we are small’


universe, this re-ionisation process was very patchy – galaxies don’t really form uniformly, they tend to be clumped together. The first galaxy to form is going to affect its neighbours, but it will do absolutely nothing to those that are further away.’ The ‘temperature knob’ approximation is therefore completely wrong, he says, but simulating this re-ionisation process can hugely increase the computation cost. ‘It’s more [computationally] expensive than every other piece of physics in the simulation combined, but it is absolutely critical and absolutely worth doing. Prior


others doing similar work is that of load imbalance: ‘When you have small regions where lots of stuff is going on, and huge regions where lots of stuff is going on, you have what’s called the load imbalance. If I was to take my cube, which I’m saying is the universe, and if I break it up into even-sized pieces and hand one piece to each of my processors, then 99 per cent of the processors do nothing, and the other one per cent are completely overloaded. We’ve had to come up with cleverer ways to distribute the work between processors. Dealing with the scaling issue is one of our biggest problems right now. How do we take our simulations and break them up, so that when two grids are next to each other in the physical universe, they should be on the same node of the computer, so that every time they have to interchange some information they don’t have to talk across the whole machine room.’ O’Shea admits that the process of scaling up is not unusual in HPC, but he is quick to point out that Blue Waters is unlike any


SCIENTIFIC COMPUTING WORLD AUGUST/SEPTEMBER 2010 33


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48
Produced with Yudu - www.yudu.com