Shawn Strande discusses status of NWSC-2 procurement process

I am the manager of the High-end Services Section (a.k.a., HSS), which is the group responsible for operating NCAR’s high performance computing, storage, and archive systems and services. HSS is composed of about 15 expert systems architects, developers, engineers, and students who are dedicated to providing our users with the most stable and reliable high-performance computing environment possible. 

One of the primary reasons I joined NCAR was the opportunity to lead CISL’s next-generation HPC procurement, NWSC-2, so called because it will be the second HPC system to be deployed at the NCAR-Wyoming Supercomputer Center (NWSC). Large-scale HPC procurements, like NWSC-2, give us a unique opportunity to assess the current and future trends in supercomputing, and translate these developments into a system that will advance the computational science work of our users. Yellowstone ushered in a new era of peta-scale and data-intensive computing for NCAR and saw the opening of the world-class NWSC data center in Cheyenne, WY. By nearly every measure, it has been a tremendously successful effort, enabling new science, and giving our technical teams an opportunity to develop the processes and skills needed to support petascale computing.

For NWSC-2, we are facing a very different challenge. Historically, with little or no optimization work, our applications have enjoyed substantial performance improvements on a per core basis, from one system to the next, thanks to increases in clock speed and other design enhancements. In the timeframe of NWSC-2, we expect that the core speed in future processors will deliver no such benefit and thus the “free ride” for performance will come to an end. Future performance gains are expected to come primarily by exploiting significantly more on-chip parallelism, using techniques like Single Instruction Multiple Data (SIMD) vectorization and threading. HPC vendors are now designing the systems to support this computing modality, and we expect these to be available in the timeframe of NWSC-2. The upshot is that we expect the NWSC-2 system to support both our current production computing workload, and to enable the application refactoring efforts required to make use of future generations of computer architectures. The good news is that with such efforts, we are laying the foundation for significant gains in performance, and thus the scale and complexity of the simulations these systems will support.

Over the past few months a small team of us have been doing the initial scoping and planning work for NWSC-2.  These are some of the key elements and outcomes from the procurement process thus far:

  • A Science Requirement Advisory Panel is providing insight into our community’s science objectives and computational requirements for 2017 and beyond.
  • We expect the NWSC-2 system will be composed of primarily traditional processors, but with a significant component of next generation “many-core” processors that will allow model development teams to begin the work of refactoring their applications.
  • We are mining performance and operational data from the Yellowstone environment to provide insights about how NWSC-2 should be designed.
  • We have recently released a Draft Technical Specification (DTS) to assist those companies who may be interested in submitting a proposal. The DTS will become a key document in the formal Request for Proposal process.
  • We plan to release the Formal Request for Proposals in early 2015.
  • We are working closely with model development teams to improve application performance on Yellowstone with and eye towards future architectures expected in the NWSC-2 system.
  • We expect the NWSC-2 system to go into production in January 2017.
  • The operation of Yellowstone and NWSC-2 will overlap through 2017, and Yellowstone will be decommissioned in December 2017, after five years of service.

I invite those interested in learning more about NWSC-2 procurement to visit