Allan Torres, Founder, The Torres Group, LLC--
Cray Inc. is listening to its customers about their “Pain Points,” says Cray President and CEO Peter Ungaro. A presentation by Ungaro is usually an open and relaxed talk interspersed with humor, interesting insights, and a long-term view. He did not disappoint attendees of the 8th LCI International Conference on High Perfomance Clustered Computing. The focus of his May 17 keynote was “From BeoWulf to Cray-o-Wulf: Extending the Linux clustering paradigm to supercomputing scale.”
Ungaro presented market realities and facts--like supercomputing with commodity processors will become almost solely focused on scalability, and that clock rates are not increasing. But the proliferation of multi-cores has renewed interest in novel processing architectures and accelerator technologies.
Cray is listening to its customers about their “pain points,” he says: that big clusters are hard to use and maintain related to power, cooling and floor space; lack of interconnect performance, applications, and programming; reliability, serviceability, and availability (RAS); storage and data management, and multi-processor and accelerator support.
Ungaro says Cray is bringing its Big Iron high-productivity computer systems approach and expertise to the commodity-driven cluster market, with a number of new and innovative directions and insights all based on their premise that pure commodity clusters begin to break down in a number of ways after about 1,000 processors. This approach comes from the direction of a convergence of a number of technologies and capabilities that Cray has expertise in or is developing with their Adaptive Supercomputing Framework, a very tightly-coupled integration of hardware and software that is focused around a building block architecture. This allows flexibility to a diverging market, where the high-end users always need more compute power and have different needs than mainstream users in the low and mid-range of the market. Making clusters easier to use is expected to draw in a new user set, especially if the needed parallel applications exist.
So what path will Cray take? One predicated on many options, notes Ungaro, based on Adaptive Supercomputing by combining multiple processing architectures into a single, scalable system. The HW foundation is comprised of interconnect, file systems, storage, and packaging for which Cray is well known. On top of this scalable HW foundation will be multiple commodity and specialized processor technologies based on Scalar X86/64, vector, multithreaded and exotic HW accelerators. So the diverse user community will have the flexibility and options to choose the processors or combinations that the user will need to attain a greater portion of sustainable performance on their applications, not someone else’s benchmarks. To further optimize Cray’s Adaptive Supercomputing model, above the HW level will be the requisite ultra-lightweight Linux OS, adaptive software such as libraries, tools, compilers, and runtime environment. All of this leads to a transparent interface to the application level and user.
When asked where will future performance gains come from, Ungaro responds, “tight integration of all of the various components of a system to enable scalability and tools to help the user take advantage of the architecture.“
Ungaro says Cray sees its market position “on the high-end … with systems in the 5K-10+K socket range, however, we are building systems targeted at 1,000 sockets and lower when it makes sense. But we see our technology providing the most value to customers at 1000+ sockets, at the high-end of HPC.”
Allan Torres is founder of The Torres Group, LLC and an independent consultant in the high performance computing, networking and storage markets. Contact him via http://oftheuniverse.com/index.html and read “About ” related to the Seymour Cray Documentary and “Sponsorship Opportunities” for SC’07.