On Saturday, November 12, 2005, I went with a group of about 20 people from the Scalable Computing Laboratory (SCL), a part of Ames Laboratory at Iowa State University, to Super Computing 2005 (SC|05). We left Des Moines airport at about 12:15PM. We were concerned about the weather, and in fact there were tornadoes a few hours after we left, but our flight to Minneapolis went fine. We had around a 4 hour layover in Minneapolis, which was generally spent trying to fine‐tune some of the programs that would be on display at the conference. Finally, our flight left and we got into Seattle around 7:30PM. After checking in to the hotel, we explored the city a bit and had dinner at McCormick’s fish house before going to sleep.
On Sunday and Monday the students in our group were enrolled in the tutorial program at the conference. On the first day, I attended the “Parallel Computing 101” tutorial, which was conducted by Dr. Quentin Stout of the University of Michigan and Dr. Christiane Jablonowski at National Center for Atmospheric Research. Although the Scalable Computing Lab (SCL) is mainly focused on cluster computing, I hadn’t done any actual parallel programming, and wanted a thorough overview of the issues involved in this.
On Monday I attended parts of several tutorials. First, I attended part of the tutorial on the Common Component Architecture, which is designed to allow improved re‐use of code between scientific computations and to allow easier creation of new programs using already‐existing modules. Part of the tutorial was a “hands‐on” session and I did not have a laptop with me, so I attended other sessions. I listened to a talk on the future of supercomputing by Dr. Peter Kogge at Notre Dame. He discussed the International Technology Roadmap for Semiconductors and its impact on the future of computers, including supercomputers. I also attended a tutorial on programming in OpenMP. OpenMP is a set of compiler directives and library routines to easily convert an existing serial program into a parallel program for shared‐memory computers.
Following the tutorials, there was the opening gala for the convention at 7PM in the exhibit areas. During the gala, I was at the SCL booth talking to people who came by about the research projects at the lab. There were over 220 exhibits at the conference, mostly from hardware providers, such as IBM, Sun, and Intel; and government or university labs, such as Brookhaven National Lab and Oak Ridge National Lab. Various foods were available during the gala, and the bar there featured a drink created for the conference called the supercomputini. I left the alcohol to those over 21.
In general, simple decision support applications do not need the capabilities of supercomputers. Large databases requiring large amounts of storage and data mining programs or large simulations can use the computational power, but in general the computational power requirements are not excessive. Many decision support systems are designed to run on a desktop computer or on a low end server, and do not need state‐of‐the‐art computational power. The conference was not particularly targeted to companies that would use supercomputing solutions in their business and there were few exhibits from software vendors; Microsoft was one of the few large software vendors with a significant presence at the conference.
On Tuesday morning, before the keynote speech, several of the people responsible for the conference gave speeches as well. Bill Kramer, the Chair of SC|05, gave a summary of the conference. There were over 9250 registered attendees for the conference this year, an increase of over 500% over the past 10 years. There was also information from the Chair of next year’s conference, as well as speeches from the heads of ACM and the IEEE Computer Society, the sponsors of the conference.
Bill Gates delivered the keynote speech titled “The Future of Computing in the Sciences”. After summarizing the current state of the computing industry and software, he discussed the future of computing. As clock speed increases are declining, parallelism through multi‐core and multi‐processor computers will need to be used to continue current trends of performance increases. Gates suggested that a “personal supercomputer” is the wave of the future, where such computers are available for under $10,000 and can be in a research lab locally, while larger jobs are sent to a more powerful computer elsewhere. Considering that our desktop computers of today are more powerful than the supercomputers of 20 years ago, this seems like a reasonable prediction. He also sketched a model for technical computing. The model starts with collecting data from sensors, models, or other computers. Data is stored on the network, and then people can access the data and do analysis. He gave an example of earthquake data being stored with a research paper, and people who are reading the paper being able to examine the program used and run it themselves to see how the results are obtained. This obviously uses quite a lot of decision support capabilities. The user interface in the slides looked very nice, but it will take a lot of effort to get the capabilities shown in the mockups to actually work.
During Gates’ speech an example with distributed Matlab was demonstrated. Kyril Faenov, Director of High Performance Computing at Microsoft, took a data set of protein spectroscopy readings from cancer patients and non‐cancer patients, and used a genetic algorithm distributed over 320 processors in multiple locations running multiple OSs to try to determine which proteins best distinguish between cancer and non‐cancer patients. Once again, the analysis system looked easy to use, but it will take time to get a system set up where people will easily have access to supercomputing power. There were some amusing questions for Bill Gates after the speech, including somebody who asked if he would talk at the grid forum, and one question asking that given the energy requirements of the chips, what he thought about the energy problem.
Following the keynote, there were technical sessions on various supercomputing topics. As the talks later in the week appeared more interesting, I took the day to explore the convention floor. The vendors had many presentations about their products, and most had T‐shirts to give away or drawings for free iPods for the people who listened to their presentations. There was also a guy personally giving out plastic antennae, which were amusing. After dinner, I went with a group from the SCL to a party thrown by Microsoft at the Seattle Center by the Space Needle. We took the monorail to the center and got there around 9PM. The feature was a concert by Sheryl Crow, which none of our group was particularly interested in. The Science Fiction Museum and the Experience Music Project were interesting though. They also had another invented drink, the “Cluster Revolution”. I was talking with people from the various labs and companies there, including two former Iowa State students who were now with the Krell Institute.
Wednesday had several technical sessions related to decision support topics. In the Masterworks session titled “The Strategic Future of Data and the Mining of Massive Data Sets”, Robert Grossman of the National Center for Data Mining and Usama Fayyad of Yahoo discussed the challenges that arise from having large sets of data and how to analyze it using data mining solutions. Following the talks, I had my final hour working at the booth. I then went to lunch with a group from the SCL at a Thai place a few blocks from the convention center. When I got back, I listened to a talk by Dr. Mark Gordon, the head of the SCL, on the computational chemistry software used by his research group. There was a party hosted by IBM which was supposed to be very nice, but invitations were limited and I did not get one. Thus, I had dinner at the Cheesecake Factory and started packing and watched TV that night.
Thursday morning I finished packing and went to the convention center. I attended a discussion on performance prediction and analysis, which discussed various tools to predict the performance of various different supercomputers on programs and using the available resources best based on this information. In this setting, decision support used to determine how a supercomputer should be best used.
Following this session, I had to leave the convention to catch my plane back to Des Moines, Iowa. The 5 of us on that flight took a limo from the convention center to the airport. Our plane from Seattle to Minneapolis was running a bit behind schedule, causing us to get to the food court just after most of the food vendors in the airport closed. Due to the cold weather and de‐icing procedure, our plane to Des Moines was late. We arrived in Des Moines around midnight.
Editor’s note: Alexander Power has been the Webmaster at DSSResources.com since 1999 and is currently completing his senior year at Iowa State University in Ames, Iowa with a degree in Math and a minor in Computer Science. This trip was funded by Ames Research Lab where Alex has been a student research assistant since May 2005.
Author: Alex Power