From the video A New Type of Research
Dr Arthur Trew, EPCC Director: Computational simulation is not a technology fad. What we are doing is to reproduce in the 21st century what they did in the Renaissance. What they did at that time was use state-of-the-art technology - it was optics - to take science forward. Here we are using state-of-the-art technology - the biggest fastest computers in the world - to develop a new type of research.
AT: What we want to do is use these computers to investigate areas where the traditional theory and experiments simply don’t work. Where the problems are too big, too small, too fast, too fleeting or simply too expensive for the more traditional approach of calculating what happens with a pencil and paper or trying to study it in a laboratory. This approach is already paying dividends in a wide number of applications in both academia and industry. For example, we are working with researchers who are studying the structure of sub-nuclear particles. At the other extreme, those investigating the structure and evolution of the universe. In terms of our industrial links - and these are vital to EPCC in succeeding in its technology transfer mission - we have reapplied techniques that we have developed in academic research to a wide range of businesses. From modelling communication networks for Cisco through to improving the effectiveness of Arran Aromatics’ production line. In my view, computer simulation has still a very long way to go and EPCC is looking forward to being in the vanguard.
New Material Structures
Professor Mike Cates, School of Physics: What we do with computers is design, and try to understand, new types of material. These are based on colloid particles which are dissolved in various different kinds of fluids. One case we are particularly interested in is where you have two fluids which at high temperature are miscible and then if you drop the temperature they start to de-mix. The colloidal particles are engineered so that they prefer to sit in the interface of the two fluids. As the fluids de-mix they collect the particles and create, we believe, a new kind of gel structure in which the colloids cover the interface and are jammed up to create a two-dimensional solid film.
MC: This kind of structure is completely new. We discovered it first by computer simulation but now are exploring it, creating the exact structure in the laboratory. The advantage of computer simulation is you can change the particle properties, you can alter the colloids, you can change the disquality of the solvents in all kinds of ways, where as if you did this in the lab would take years and years to achieve. The basic difference between, in our field, what you can do with an ordinary set of workstations (the type of thing you can get on a stand-alone research council grant) and the kind of facilities which are now being put in place at the ACF, is the difference between being able to do a simulation in two dimensions as a kind of toy model and fully three-dimensional realistic simulation.
MC: That’s the difference, that’s the extra powers of ten in computer power that we need.
The Future of Computation
Professor Richard Kenway, School of Physics: While machines come and go, software persists. The intellectual energy that we put into science is embodied in our software, which are becoming more complex and of much greater value to our science. So I think as time goes on we are going to have to put software first. That means either we will increasingly design our computers to fit the software or that our software shall have to be designed to ride the commodity mass market wave, so that we can exploit the mass market to do our high-performance computing.
RK: At the ACF we are exploiting both of these routes so we have at the moment QCDOC, which is a specially designed machine to achieve very high compute intensity at a relatively low power in order to tackle a very challenging problem in particle physics. But equally we are building up clusters of commodity PC-based systems which enable us to produce very cost-effective high-performance computers for quite a wide range of applications.
Biology and Genetics
Professor Graham Bulfield, Head of College of Science and Engineering: In biology we have been in an intensely reductionist phase. Getting down to the level of the genes, sequencing the genes and finding out what the genes begin to do. Of course we then have to go from the gene right the way through metabolism and development up to the characteristic itself, such as growth or a particular disease or so on. This means that, instead of having a reductionist phase, we have to go from the gene through complexity and the complexities of metabolism and development are so great, that we have to start using high-performance computing of various types.
HPC at the University of Edinburgh
Professor Timothy O’Shea, Principal: The University has a strong reputation around the world for its scientific work and an equally strong reputation for its work in informatics. The place where these two domains intersect is in high-performance computing.
TOS: High-performance computing builds on the very major computer science strengths of the University and is entirely necessary for modern scientific work in physics, in chemistry, in a whole wide range of domains. That is why I judge it so important for us to support the Advanced Computing Facility so that we can continue to maintain our lead in high-performance computing.
Real Time Prediction of Fire Growth
Professor Jose Torero, School of Engineering and Electronics, Firegrid Network: The building of fire implies millimetre-scale flames, for example, all the way to metre-scaled structures and all the way up to kilometre-scale pollution problems. So you need extensive computational resources to be able to actually work from that very, very small scale all the way up to the larger scales. The problem with that is if we run, for example, this type of computation in the case of an emergency and we are trying to help, for an example, the fire brigade, these computations would take weeks - or months sometimes - to run. Therefore it is absolutely necessary to be able to use, on a need-basis, deployable resources via the Grid to try and reduce the computational timescales into a mechanism by which we can forecast the behaviour of a fire in the case of an emergency.
Drug Discover Techniques
Malcolm Walkinshaw, School of Biological Sciences: The basic aim is to design new drugs for all sorts of different types of diseases. From parasite infection, cancer, inflammation… everything. So now we have this vast amount of information on sequences of proteins and we are converting them here into three-dimensional structures so we can clone, express, generate proteins of any potential drug targets. We identify a drug-able target, its experimental structures, and then we go into computational biology where we use these three-dimensional structures as templates so that the protein targets will probably have small pockets or holes or enzyme-active sites and we can use these to design specifically small molecules to block these sites and block the activity of the protein.
MW: The big excitement with these new computational facilities is that we can do this on a massive scale. So up until now we have looked at one target and a few hundred small molecules. Now we can expand that to look at all the proteins in the human genome or in the parasite genome and try and match these structures with ten to the twelve, ten to the fifteen potential small molecules. If we can use computers like Blue Gene to speed up that process, then the end of the line is going to be new, interesting small molecules that can be tested biologically.
Geological Fault Simulation
Professor Ian Main, School of Geosciences: I am interested in how things break in general and this is an example from a sandstone. You can see the different formations localised onto these sheer bands and they show up as kind of proud of the rest of the rock, which is eroding more quickly because it is more porous than the bits that are broken and where all the material is crushed.
IM: So the question is: How does deformation localise in the Earth’s crust? What we have done is constructed simple numerical models to try and reproduce that behaviour. One example is the simulation of the growth of a mega fault, for example The San Andreas Fault. This animation shows the evolution of faults from initially a distributed array of active faults, gradually with deformation concentrating more and more on the dominant mega fault, which might be The San Andreas Fault or another plate boundary fault. This is exactly what we see in nature. What we would like to do with the new facility is to take that two-dimensional picture that we have been able to build up to date and turn it into a three-dimensional version, which we can actually use for practical applications.
FPGA High Performance Computing
Dr Mark Parsons, Commercial Manager EPCC: Field Programmable Gate Arrays or FPGAs are becoming more interesting to the high-performance computer community. They are a type of silicon chip that allows you to express any type of algorithm or other computer processor on the bare silicon. The challenge is that they are much more hard to programme.
MP: The FPGA High Performance Computing Alliance shall be siting their machine out at the ACF. This is a new two-year project funded by Scottish Enterprise to develop two things. One is to look at how we take forward the FPGA community and our expertise in Scotland and develop the knowledge base we have in Scotland in this area. The other is to build an FPGA supercomputer which will demonstrate everything we learn during the process. EPCC is always interested in looking at the forefront of computing technology and the reason we are interested and excited about this project is that we want to understand how we use FPGAs and whether they are cost effective.
Complex Biological Systems
Dr Jason Crain, University of Edinburgh - T J Watson IBM Alliance: A major challenge in the life sciences today is to understand how biological functional structures are formed. In other words what are the design principals and how do these assembly mechanisms go wrong and cause disease like Alzheimer’s and CJD? To address some of these fundamental questions, scientists from the University of Edinburgh and IBM Research in New York have teamed up to combine extreme supercomputing technology, like Blue Gene, with new simulation and computational methods with novel experimental techniques to study these processes at atomic and molecular detail. An initial aim of these studies is to understand how complex structures emerge from simple molecular building blocks and then to scale up these models towards real biological systems. At every stage our objective is to build confidence by validating the computational models against experiments.