Inside this Issue

• Global Tsunami Model
• Alaska Weather Modeling
• Geochemical Interfaces
• VRcussion
• Clusters at ARSC
• Intern Summer Research Program
• Visualizing Frost Boils
• Permafrost Visualization

Dr. Greg Newby, Acting Chief Scientist, ARSC

Story by Greg Newby

It is a great pleasure to look back on the scientific accomplishments ARSC has engaged in over this past year, and to look forward to the future. The biggest highlight of the year might be the variety we have seen, and the growing scope of scientific activities at ARSC. In this review, I will look at some of the science that ARSC has been most effective at in this past year, and also provide a preview of some of our goals for the upcoming year.

In the supercomputing world, most experts agree that today’s biggest supercomputers are good at some types of computational problems, but not all. Formal models, notably Amdahl’s Law, place limitations on the amount of speedup a particular computational problem can get by adding more parallel processing power.

At ARSC, our biggest computers spend most of their day working on problems for which they are well-suited. Two big classes of these problems are computational fluid dynamics and finite element analysis. Several computational applications with these characteristics are highlighted in this issue, including weather, tsunami modeling and surface chemistry. (See Challenges, pages 3, 4 and 6.)

With scientific programs (nicknamed “codes”) in these areas, the domain under study is divided into many different subcomponents. For example, a weather model might divide a geographic region into some number of cells, based on latitude and longitude. In computational chemistry, the interactions among different atoms and molecules might be modeled. By dividing a problem into many semi-independent parts, each part can be computed independently – or at least independently until it’s time to get input or provide output to an adjoining part (like when wind passes over a terrain feature, changing the temperature). If the independent parts can make good computational progress before needing to communicate, then our largest parallel computers might be a great fit – by dividing out the work onto multiple CPUs. But as in life, putting things into practice can be pretty complicated. For example, many computational problems scale to a point, then degrade. Perhaps using 16 processors gives vastly improved performance over just one processor, but adding another 16 may not help. It’s not easy to use these large systems, but the potential payoff is that we accomplish science that can’t be computed otherwise.

ARSC activities are not limited to those computational science challenges best suited for our largest computers, however. Because ARSC supplies a wide range of systems, as well as sophisticated storage, networking and visualization facilities, many other types of science can be supported. Many of the intern projects, as well as the visualization aspects of FrostByte and the frost boils analysis, fit this profile. (See Challenges, pages 18 and 20.)

In the future, we can expect this diversity of research challenges and the systems, software and staff that support them to grow. Thanks to the continued emergence of standards and enhanced capabilities in computers, it’s easier than ever to apply “big iron” to a science area. These trends aren’t just about MPI and the latest Fortran standard, but also include the Linux operating system, new programming languages, such as Unified Parallel C (UPC), and software packages, such as MATLAB® and IDL, that pack a lot of complexity into just a few lines of code.

Because today’s computers are so fast and have so much memory, “big iron” isn’t what everyone needs. The middle-sized servers at ARSC offer huge potential speed-ups over desktop workstations, but are still a fraction of the size of our largest high-performance systems. We’ll be working over the upcoming year to offer more of these “right-sized” systems, with the software that scientists need. For some scientists, these systems and software will provide much-needed platforms for pre- and post-processing of data. For other scientists, these systems might actually offer a better fit than our largest supercomputers. One general class of science that might not be a good fit for our largest systems is known as “data-intensive computing.” With data-intensive computing, the emphasis is more on data movement or processing than on computational power. Big databases, image repositories, genetic sequences and digital libraries are examples of data-intensive computing. ARSC is working to grow our capacity to better support these types of data-intensive science applications, both through more systems and software, as well as by adding to our staff expertise.

The good news about all of these plans is that it’s relatively easy to add new systems and new software at ARSC, since we already have an extensive infrastructure and highly trained staff. Supercomputing centers are, by nature, equipped to handle the changes in scientific uses of systems. But growing capacity can’t be done randomly, or there’s a good chance of missing opportunities or developing solutions that don’t match users’ needs. For this reason, ARSC spent time over this past year gathering feedback and input from new and potential users.

Some of the outcomes from this information gathering are already evident–including some new Linux-based systems, on-site training for MATLAB® and changes to batch queue structure and limitations. Others will be rolled out over the upcoming year, and include smaller systems targeted at pre- and post-processing of data and the use of computational packages, such as ABAQUS and FLUENT. We’re also looking to provide Web/database systems for ARSC users, as a method for tracking their computational runs, and distributing results to others.

In all, it has been a very exciting year at ARSC. We have continued to serve our core computational scientists, and have had a very stable and productive year for Klondike and Iceberg, our largest systems. It has also been a year of introspection and growth, during which the ARSC staff engaged in self-study and evaluation, leading to positive changes in how we work. ARSC is taking significant steps to broaden our activities and to reach out to new users while maintaining the high-quality service that existing users expect.

I hope you enjoy reading about some of this past year’s highlights. The projects described here are innovative, and highlight different aspects of ARSC’s computational science capability. Moving forward, you can expect continued variety in the computational science activities that ARSC is engaged in. You are welcome and encouraged to get involved with ARSC’s continuing evolution.