AUTOMOTIVE

SCO: Last week Dell hosted an event entitled "Commodity, High-Performance Cluster Computing Technologies and Applications," at the SCI 2002 conference. Please tell the readers how it went. What are a few of the highlights in your opinion? ROOHOLAMINI: Dell’s “Commodity, High-Performance Cluster Computing Technologies and Applications” event focused on the use of commodity, off-the-shelf components for building high-performance computing clusters (HPCCs) to solve technical computing problems. The event was well attended, with international participants from academia, national laboratories and industry. Several Dell partners including Oakridge National Laboratory, Cornell Theory Center, University of Texas Department of Petroleum and Geosystems, MPI Software Technology Inc. (MSTI), Cray Inc., and the University of Houston presented papers exploring best practices and the hardware and software building blocks used for architecting HPC clusters. Attendees also participated in a panel discussion that generated excellent debate on a variety of timely issues including programming models for HPCC, grid computing, total cost of ownership (TCO) issues, performance vs. management, and the use of high-speed interconnects such as InfiniBand. Of particular interest was a discussion surrounding TCO issues. Participants agreed that the initial acquisition cost of the HPCC solution should not be the determining criteria when an organization is selecting a vendor. Often, organizations focus on the price of the systems and not the total package including support, services, scalability, etc. These organizations usually end up paying more in the end. SCO: What do you consider to be one or two of the most significant advances in building high performance computing clusters for complex research? ROOHOLAMINI: The enhanced price/performance ratio of Intel-based platforms is one of the most critical and important advances in architecting HPC clusters. When used as the building blocks for any complex computing system, these platforms offer excellent performance, reliability and scalability at an attractive price. In addition, Intel-based platforms’ small footprint and superior manageability make them ideal for deploying in a large-scale HPC cluster. SCO: You gave the keynote at SCI entitled "Trends in High Performance Computing." What are a few of the trends outlined in that presentation? ROOHOLAMINI: In my keynote, I presented a series of trends I’d observed in the Top500 ranking of supercomputers (www.top500.org) recently released by the University of Tennessee and the University of Mannheim, Germany. HPCC has served as a disruptive technology in the traditional supercomputing space. Since the advent of HPCC, the price of one GFLOP of computing power has continuously declined. Today, nearly 780 GFLOPs of reliable and scalable computing power can be purchased for close to $1 million. This price includes substantial services and support. In the presentation, I noted that the definition of “commodity” has traditionally only referred to hardware. However, software and services have now also become commoditized, and this has made significant impacts on the industry. I also addressed the fact that the line between technical and commercial computing continues to blur. The convergence of these two traditionally distinct computing disciplines is evidenced by the fact that features and functionality inherently demanded by one are starting to be mirrored by the other. For instance, Dell’s HPCC customers are increasingly demanding the integration of a shared, reliable and scalable secondary storage infrastructure within their HPCC. Storage technologies such as Storage Area Networks (SAN) and Network Attached Storage (NAS), traditionally associated with commercial computing, are being integrated within HPCC environments at a growing rate. I also spoke about developments made by the Open Source Cluster Application Resource consortium (OSCAR). OSCAR is a set of tools and utilities designed for the deployment and operation of a HPCC. Dell is one of the founders and voting members of this consortium. SCO: Any interesting recent customer news you'd like the readers to know about? ROOHOLAMINI: Dell’s HPCC program continues to gain customer momentum. Johns Hopkins University’s Department of Earth and Planetary Sciences is using its Dell cluster to perform ocean circulation modeling at its Baltimore campus. The cluster fuels 3D modeling of the earth and ocean to simulate weather patterns, the effects of global warming and the effects of manmade contaminants on the environment. Compagnie Générale de Géophysique (CGG), a worldwide oil company, is using its Dell cluster to do complex modeling of the earth's subsurface to help get more accurate pictures of oil deposits for oil drilling, making the process more efficient, less expensive and more environmentally friendly. The Departments of Chemistry, Physics, and Chemical Engineering at the University of Notre Dame are using a cluster of Dell workstations to study a variety of topics ranging from building next-generation computers using molecular electronics to studying the formation of galactic halos in space. The peak performance of 700 GFLOPs allows the researchers to tackle complex, multi-scale projects in the areas of nanoscience, biological chemistry, and astrophysics. SCO: Is there anything else you’d like to add? ROOHOLAMINI: There are quite a few design choices that need to be made when building an HPC cluster including the class of server or workstation, the interconnect, the operating system, the message passing middleware, and the cluster management tools. Often, each application requires a different design choice. At Dell, we believe the best way to serve the needs of our customers is to create a vast knowledge base of these technologies to build upon in order to create the optimal HPCC solution for their needs. We continuously monitor new and relevant technologies and have streamlined the processes to assess their performance and communicate the findings to our customers in an efficient manner. We believe that HPCC has now truly entered the commodity stage of its life cycle. As such, we are committed to further technology developments in the space and aggressively commoditize HPC clusters, leveraging the strengths of our direct model, strong partnerships and ability to rapidly integrate emerging technologies.