CS professors Godfrey and Wang join project building an intelligent and self-adaptive operating system

“Whenever we use a laptop or even a phone, underneath the hood, an operating system is running everything – allowing a plethora of applications to dynamically share a system, and managing memory, compute, and network communication resources,” said the University of Illinois Urbana-Champaign Computer Science professor Brighten Godfrey. “Today, that happens mostly through manually-written heuristic policies, but this is becoming difficult with complex applications running in complex environments, like compute clouds and robots. Could an intelligent and self-adaptive operating system do better?”

This is the crux of a ground-up effort to build a new way of operating systems for computers.

Godfrey and CS colleague Gang Wang are principal investigators in a five-year, $12 million research project funded by the NSF that aims to harness artificial intelligence to boost the performance and energy efficiency of computer operating systems.

NSF Expeditions in Computing: Learning Directed Operating System (LDOS) – A Clean-Slate Paradigm for Operating Systems Design and Implementation will be led by a team from the University of Texas at Austin with lead PI Aditya Akella, Professor and Regents Chair of Computer Sciences at UT Austin. In addition to Godfrey and Wang, the team includes computer scientists from the Texas Advanced Computing Center, the University of Pennsylvania, and the University of Wisconsin-Madison, with industry partners from Amazon, Bosch, Broadcom, Cisco, Google, and Microsoft.

Wang recalled his involvement began when Akella told him of “a vision of building a brand new operating system that is completely machine learning based.”

Wang, whose research topics include security, said, “When you build a brand-new operating system, you're going to have security problems. You're going to have privacy problems to address, and that's where I feel that my research can come in and contribute. There are a lot of new problems related to the intersection of systems and AI that make me super interested.”

Today’s operating systems often have rigid rules for allocating hardware resources between different applications running simultaneously. This inflexibility prevents the integration of new advancements, leads to poor performance and inefficient use, and poses a significant barrier to computer hardware and application innovations.

Wang described the LDOS goal: “We're going to tear down all the existing barriers and start to build something that learns from the data and adapts as the environment changes or as the need changes. That's a really exciting vision. It allows you to build new applications or capacities that are not easily achievable in the past model.”

Godfrey’s research group at Illinois had previously investigated applying learning approaches to systems components such as congestion control, where the OS needs to make quick decisions about how fast to send data. “Learning-based techniques helped there, because simple heuristics, which make simplistic assumptions about the network environment, can lead to poor performance.  Similarly, an LDOS will try to figure out the right way to manage its resources. But we will have to solve some very challenging problems – like how to coordinate many learning-based agents acting on different components of the system simultaneously, and how to ensure consistent, dependable performance.”

The NSF Expeditions in Computing program was established by the NSF’s Directorate for Computer and Information Science and Engineering (CISE) to build on past successes and provide the CISE research and education community with the opportunity to pursue ambitious, fundamental research agendas that promise to define the future of computing and information.

The LDOS project will include instructional and outreach elements, creating undergraduate and graduate curricula consisting of modules, courses, and certificates. Initiatives to broaden participation will cultivate leadership skills among underrepresented groups in AI and prepare them for careers in AI and computer systems technology and research.

Godfrey will be involved in the effort and said, "One of the big goals is to draw more students into systems research, which we think is hugely important. Systems are underlying everything we do, and every time there's a new technological development like AI, we need new systems to help realize it. We want to make people aware of these interesting, challenging problems.”

“The project will include summer institutes for students, including pre-college and undergraduates, that will help draw them into this world of systems plus machine learning research. And my regular courses here at Illinois will also certainly be influenced by the work in this project,” he said.

Wang added that “This is going to be a pipeline, so to speak, for a cohort of high school students to undergraduates to graduate students and even postgrads. They will attend summer schools, and they will participate in research. Curriculum development is an addition that integrates research with teaching and student training. I think we have a lot of students in AI, but we need more students in systems and AI who have this joint knowledge and expertise to do great things.”

The NSF Expeditions program is highly selective. This year, three projects were funded with $36 million. LDOS is the second Expeditions project to feature Grainger College of Engineering faculty. In Fall 2022, NSF awarded a seven-year, $15 million project to a multi-disciplinary, multi-university team led by The Grainger College of Engineering. Mind in Vitro—Computing with Living Neurons will imagine computers and robots that are human-designed but living. Mind in Vitro was one of two projects funded that year.