The status of exascale computing in the US

29 Jul 2015 Washington - At a meeting of the Advanced Scientific Computing Advisory Committee (ASCAC) of the Department of Energy (DoE) on July 27, the status of the US exascale initiative was reviewed. Steve Binkley Associate Director, DOE Office of Advanced Scientific Computing Research, concluded in his presentation that the next generation of advancements will require Extreme Scale Computing of 100-1,000X capabilities of today's computers with a similar physical size and power footprint. This poses significant challenges for power consumption, high parallelism, reliability. Extreme Scale Computing cannot be achieved by a "business-as-usual", evolutionary approach. Hence it is necessary to initiate partnerships with U.S. computer vendors to perform the required engineering, research and development for system architectures for capable exascale computing. Exascale systems will be based on marketable technology; it will be not a "one off" system. They must be productive systems, usable by scientists and engineers.

The "Report From The ASCAC Subcommittee on EXASCALE" was presented by Dan Reed, chair of the ASCAC. According to Reed, the subcommittee strongly endorses the DOE plan for exascale computing development and deployment. But Exascale computing is not a destination, but the next milestone in a journey. Advanced computing's benefits are broad and deep. The committee has formulated seven detailed recommendations:

  1. Develop a detailed management and execution plan that defines clear responsibilities and decision-making authority to manage resources, risks, and dependencies appropriately across vendors, DOE laboratories, and other participants.
  2. As part of the execution plan, clearly distinguish essential system attributes (e.g., sustained performance levels) from aspirational ones (e.g., specific energy consumption goals) and focus effort accordingly.
  3. Given the scope, complexity, and potential impact of the ECI, conduct periodic external reviews by a carefully constituted advisory board.
  4. Mitigate software risks by developing evolutionary alternatives to more innovative, but risky alternatives.
  5. Unlike other elements of the hardware/software ecosystem, application performance and stability are mission critical, necessitating continued focus on hardware/software co-design to meet application needs.
  6. Remain cognizant of the need for the ECI to support for data intensive and computation intensive workloads.
  7. Where appropriate, work with other federal research agencies and international partners on workforce development and long-term research needs, while not creating dependences that could delay or imperil the execution plan.

Ad Emmen