A new National Science Foundation initiative aims to develop a framework for advancing the next generation of scalable systems and applications, including HPC platforms.

The nearly $90 million program, Principles and Practice of Scalable Systems (PPoSS), will over the next decade focus on scaling systems and applications. Among the challenges is keeping pace with emerging AI and machine learning techniques that have complicated HPC design and implementation.

“Our goal here is to really advance basic research on the scalability of modern computing systems and applications,” Erwin Gianchandani, acting assistant director of NSF’s Computer and Information Science and Engineering office, noted during a recent webcast.

As Moore’s Law and Dennard scaling run their course, program officials stressed the need for new approaches.

“We are now entering a new era, where achieving the same pace of transformative computing innovation will require a refamiliarization of

different communities—algorithms, hardware, networking, software, systems—with each other’s domains and the development of new abstractions and paradigms to handle the domain-specific challenges we are already confronting,” added Rance Cleaveland, director of NSF’s Computing and Communications Foundations division.

PPoSS spans the entire hardware and software stack, and will target proof-of-principles for advanced systems, applications and tools that are both scalable and secure.

The computing effort also acknowledges the distributed nature of applications, a trend that has blurred the distinctions between advanced hardware and software. What is needed, NSF said, are “new abstractions, algorithms and system stacks.

The agency said grant proposals should describe “targeted applications, systems and platforms.” Initial PPoSS planning grants to be awarded this year will focus on forming research teams, establishing goals and proof-of-concept efforts. Larger awards will follow.

The agency will begin accepting detailed proposals for larger projects beginning in 2021. The “large” proposals will cover “full-stack” solutions that require interdisciplinary approaches. Grants for those proposals would run as high as $1 million annually for up to five years.

Program officials said they expect to fund up to 15 planning proposals. The lion’s share of the PPoSS funding—up to $80 million—is earmarked for the project phase.

Along with HPC, eligible research topics include: computer architectures, programming languages and compilers, security and privacy; theory and algorithms; and systems. “Both the use of AI and [machine learning] methods and the interplay between program synthesis and AI and ML will drive the need for more performance and more parallelism in all the research thrusts,” NSF said.

The HPC thrust specifically addresses design and implementation problems to accelerate and scale the delivery of distributed applications. A parallel track will seek to optimize “power footprints and [permit] a broad range of resource usage behaviors that may differ across components of the applications and phases of their workflows,” program officials said.

The deadline for submitting planning proposals is March 30. NSF expects to announce the first PPoSS grant awards this fall. Project grants are expected to be awarded in January 2021. Refer to the NSF solicitation for additional information.