Leadership Scientific Software (LSSw)

Leadership Scientific Software (LSSw)

Increasing diversity of leadership computing platforms

Leadership scientific computing has traditionally focused on supercomputing: building the next generation of large computers to improve the scientific community’s ability to execute larger simulations with higher fidelity, increased complexity, multiple scales and physics or related advances. The focus has been on designing and installing the next centrally-located system that represented a (virtual) destination for its users. Software design and implementation are typically part of this effort in order to explore the hardware-software design space simultaneously, assuring that the software stack and applications can leverage the new hardware.

In recent years, computing resources have diversified. While large centralized systems such as the Frontier, Aurora and El Capitan exascale systems are still essential components of the computing ecosystem, other important resources such as scientific instruments, sensors, edge devices, and cloud resources, ubiquitously available via networks as a collection of resources, are becoming critical elements of large-scale scientific computation.

With this expanded definition of leadership computing platforms, we also need to expand our definition of leadership scientific software to include a broader co-design and co-development of software products to meet the needs of scientist who use these emerging networked platforms for their research.

Expanding the definition of leadership scientific software

We define Leadership Scientific Software to be libraries, tools and related application codes that contribute to scientific discovery and insight on new and emerging computing platforms. The focus of these capabilities is to push the boundary of feasibility by exploring and developing new scientific software capabilities that both inform the design and implementation of new computing platforms and make possible their effective and efficient use.

To address our increasingly diverse and networked computing resources, leadership scientific software will need to support:

  • Continued efforts for high-fidelity modeling and simulation
  • Machine learning alongside and integrated into simulation environments
  • Large-scale scientific instruments, sensors and other edge computing environments
  • Integration of quantum computing resources

In all these situations, the term leadership scientific software means we are designing the software and the computing platforms simultaneously. Activities in this kind of dynamic environment require effective collaboration across teams with diverse skills. The goal is to address new and important scientific problems by simultaneously designing and implementing both the computing platform and software.