 |
Inside this Issue |
|
| Column: Forward Momentum in Computational Science | ||
Dr. Greg Newby, ARSC Acting Chief Scientist |
||
Story by Dr. Greg Newby
Computational science is fundamental to ARSC, and a driving force behind many developments in computer software and hardware. Defined broadly as scientific activity accomplished with, or with the help of, computational devices, computational science is applicable to many academic and scientific disciplines. Some of the main areas of computational science found on ARSC’s computer systems are at the forefront of what today’s supercomputers can do. For example, many scientists use ARSC’s systems for modeling, in which some aspect of the real world is represented in software. Rather than representing the real world in full detail, though, models need to use a simplified view. One reason for this simplification is that the world is very complex, and not fully known —so, deciding what needs to go into the model can be difficult. Another reason is that supercomputers, as fast as they are, can’t handle the full scale of computation needed to represent the domain under study. Let’s look at examples for both reasons. Consider weather. Despite folklore that tells us the flapping of a butterfly’s wing somewhere far away can impact today’s weather in Alaska, the fact is that there are any number of factors that influence the weather. The challenge is that the relative contributions of each factor, with its behavior and interaction with other factors, can be complex. Everything from surface conditions and topography to solar radiation plays some role in the weather, but how all those factors interact is not fully understood. Even if all the factors were understood, they might not be easy to represent using a programming language. Another aspect of weather is more closely related to the theoretical butterfly: scale. In the real world, we know there are very small phenomena that impact our weather. This includes things like it being cooler in the shadow of a building, and those tiny rainshowers that don’t seem to be falling anywhere else. If we want to divide the real world into some number of spatial cells, and make it easier to divide the computational work among many processors, we will rapidly encounter limitations as the number of cells grows. Computational scientists need to address these challenges through application of knowledge from one or more scientific disciplines, through detailed understanding of computer languages and how they can represent phenomena, and through practical expertise with today’s large computer systems. A Career in Computational Science Becoming a computational scientist is not easy, and if you ask several computational scientists how they came into their profession you will likely hear different stories. Although there is an element of computer science in computational science training, this is not the typical educational career path. Degrees in computer science tend to be agnostic about scientific disciplines, and have little if any focus on the challenges of large systems. More often, computational scientists have formal training in their scientific disciplines, and less formal (though still extensive) training and experience in computer science topics such as programming, algorithms, and performance analysis. For modeling, there is also a strong need for a mathematical background to develop, implement and assess complex systems. One of the most important roles for developing computational scientists is community. ARSC helps to foster this community at UAF and elsewhere through regular training and symposia, and by reaching out to computational scientists for advice on next-generation systems and software. Combining Tools and Knowledge Not all computational scientists need today’s largest supercomputers, and many do not need to spend a lot of time programming and debugging. Models, data analysis, and visualization can be accomplished on smaller systems, and often this is the starting point for someone with a computational problem to solve. For some computational problems, a desktop workstation or small cluster is sufficient—perhaps with pre-packaged software. For others, those tools are only the starting point. As the complex systems under analysis grow, and take on more characteristics of the real world, it is often necessary to look to new software or to write your own—and to run this software on ever-larger systems. Today, using large computer systems can be difficult. Despite a community of user support, most computational scientists—like all types of scientists—are addressing challenges that do not quite fit within prior approaches. To meet these challenges, scientists need to develop, implement, and evaluate their own new approaches. At ARSC, we are seeking to bring computational science to a higher level in numerous ways. As mentioned, ARSC runs workshops and seminars to help build a computational science community. There are many ARSC participants in national and international forums, helping to spread knowledge of best practices and to bring them to ARSC users. ARSC plays a role in courses on the UAF campus related to computational science by supplying instruction, computational resources, and instructional support. In the future, there could be a full degree program on campus for computational science. There are hundreds of such programs around the world, with emphasis on particular scientific disciplines or computational approaches, on programming and analysis on mathematics, and on informatics. There are already many faculty members and other supercomputer users at UAF who see themselves as computational scientists and a new degree program, center or facility would help provide a focus for their community. ARSC also fosters new users by providing access laboratories, one-on-one consulting, detailed analysis of existing computer programs or models, and discussion of how existing approaches can scale to larger systems. ARSC doesn’t draw the line at scientific modeling. To encompass all aspects of high-end computer use for science, efforts include visualization and sonification of data, large data-set processing, databases and computational portals. Sometimes computational science really is rocket science, but that doesn’t mean it’s impossible. With a community of scientists and the support of ARSC, we can all reach for the stars. |
||
| Challenges Index | ||
