Richard Moore is SDSC's Big Kahuna of Production Systems, leading the team that procures, operates, maintains and manages the center's world-class storage, data, networking and high-performance computing systems. Moore received his Bachelor's of Science in Astronomy and Applied Mathematics from the University of Michigan in 1975, and his Ph.D. in Astronomy from the University of Arizona in 1980. Moore joined SDSC in 2002.
You direct SDSC's Production Systems Division. What kinds of systems are on SDSC's machine room floor for the national user community?
Moore: We currently have three nationally allocated compute systems, all of which are specifically designed to handle the specialized computation and data-intensive needs of our user community.
Our largest system, DataStar, is an IBM Power4 machine that offers users 15.6 teraflops of power and has a number of nodes set aside for specialized purposes such as visualization and databases. In response to heavy demand by users, we expanded DataStar last fall by approximately 50% and it can now handle a single “capability” job, which uses all the machine's resources, requiring more than 2,000 processors.
SDSC hosts one of the most powerful TeraGrid clusters adding 4.1 teraflops of compute power. We also host a 200 terabyte global parallel file system called GPFS-WAN for our partner sites on the TeraGrid. And finally, we were the first academic institution to bring into production status a 5.7 teraflop single-rack Blue Gene system.
In addition to the compute systems, we provide a powerful set of data and networking resources for the user community. Right now we have 1.4 petabytes of online disk storage available – that's about nine times the size of the Library of Congress – configured in a number of file systems tailored to support large data collections.
We are currently upgrading our tape-based archival system from six petabytes of capacity to more than 18 petabytes. We host two alternative systems to access this archive – the High-Performance Storage System (HPSS) and SAM-QFS – each with its own capabilities for users. We also have a number of high-speed networks including the TeraGrid network at 40 Gigabytes per second (Gbps) and several 10 Gbps networks including Abilene.
SDSC just submitted proposals to the National Science Foundation's High Performance Computing competition. What is your strategy for high performance computing at SDSC?
Moore: SDSC is absolutely committed to high-performance computing (HPC) as the foundation for supporting NSF's cyberinfrastructure vision. Our systems are consistently in high demand by the national user community and I am certain that our proposed future systems will continue to meet the most demanding requirements of the NSF user community.
Our strategy is to continue to focus on providing balanced HPC systems that excel at supporting both compute-intensive and data-intensive applications. And it's critical to provide a capable, balanced infrastructure that maximizes the effectiveness of the computing system for user applications. I'm enthused that NSF is pursuing an aggressive initiative towards sustained petascale computing – this is an exciting challenge that I think will have great scientific pay-off.
SDSC focuses on data, so how does the SDSC production environment support users?
Moore: SDSC's data focus spans many elements of the production environment (as well as many research efforts throughout SDSC). For example, our DataStar and Blue Gene systems are tailored to support “extreme I/O” applications. This means that users can take advantage of our high-bandwidth, large-capacity parallel file systems, medium-term parking space for analyzing and visualizing large datasets, and ready access for online community data collections.
It's interesting that our historical records demonstrate a consistent trend of exponential data growth in our archives, essentially doubling every 14 months. We are upgrading our archival system to triple its capacity to more than 18 petabytes, with about five times more bandwidth than our current system. Data reliability is critical, especially as we transition from storing primarily simulation output to “non-reproducible” data. In addition to establishing tailored classes of service, we are currently working with several partners to provide geographical replication.
Looking forward, I expect that providing storage capacity and bandwidth will be relatively straightforward compared to managing huge numbers of files. Recognizing the critical importance of the Storage Resource Broker (SRB) for managing large data collections, we have recently established SRB as a production service that SDSC supports for the national user community.
While keeping up with this growth is a challenge, there are many pressures towards even faster growth. Examples of this include our DataCentral initiative in hosting large-scale data collections, the popularity of our global file system on TeraGrid, ever-faster compute systems, and an increasing focus on long-term very high-reliability storage.
Your Ph.D. is in astronomy, how does that affect the way you approach your job at SDSC?
Moore: Even though my job focuses on procuring and operating the large-scale hardware here at SDSC, I always remember that the reason we have this hardware is to enable the user's scientific advances.
With a strong background in a domain science like astronomy, I can better relate to the perspectives of our users and appreciate the science that's being created. It's a great balance. At the end of the day, the science we enable is a real source of motivation for me and my staff.
What do you do for fun?
Moore: I really enjoy traveling, whether it's discovering exciting places, getting out of the comfort zones or just kicking back. My fiancé and I spent a three great weeks in Thailand last year. It was my first trip to southern Asia. The phrase that kept going thru my head was that “it's all the same and it's all different” – seeing the commonalities that bridge very different cultures.
It's great to live right next to the ocean and terrific beaches. I'm learning how to surf but somehow I think my Midwestern background must be a limiting factor (it can't be my age).
This article was reprinted courtesy of the San Diego Supercomputer Center, http://www.sdsc.edu/.