Research that will use artificial intelligence to boost the performance and energy efficiency of computer operating systems will be led by a team from The University of Texas at Austin, thanks to a major grant from the U.S. National Science Foundation’s Expeditions in Computing program. The University of Wisconsin–Madison, a leader in operating systems and machine learning research, is one of four universities teaming up on this transformational effort. Professors Michael Swift and Shivaram Venkataraman will lead the UW team.
Today’s operating systems pose a significant barrier to a number of promised exciting innovations in computer hardware and applications — from personal assistant robots to autonomous vehicles to edge computing that could enable smart cities and less energy-intensive cloud computing. These operating systems often follow “one-size-fits-all” rules for how hardware resources get allocated between different applications running simultaneously. The inflexibility of these rules makes it hard to integrate new advancements, resulting in poor performance and inefficient use, a problem the research team plans to tackle by leveraging AI.
“Our project will employ AI-aided intelligent resource management and auto-adapt as new applications and hardware emerge,” said Aditya Akella, the Regents Chair in Computer Sciences #1 who is leading the project. “This will enable computing devices to be used at near-optimal efficiency while meeting the needs of arbitrary applications, and it will make computing infrastructure ‘self-driving’ by automating OS implementation and management.”
Akella said the project would extend beyond academics, bringing together not only computer scientists from UT, the Texas Advanced Computing Center, the University of Illinois at Urbana-Champaign, the University of Pennsylvania and the University of Wisconsin–Madison, but also industry partners from Amazon, Bosch, Cisco, Google, Microsoft and Broadcom.
“These partners collectively develop and run operating systems for much of the world’s computing infrastructure, and we will work with them to create the next-generation open-source intelligent and adaptive OS,” Akella said. “We believe that the project offers a timely opportunity to fundamentally change the trajectory of computing.”
He added that the new style of OS could help autonomous robots become “the smartphones of the 2030s and beyond.”
“The smartphone revolution was enabled by new OS frameworks (for example, iOS and Android) that enabled users to run arbitrary third-party apps that have come to determine how they interact with technology and essential services,” he said. “LDOS can similarly usher in the era of affordable personal robots and apps that support our day-to-day activities and improve society for all, particularly for aging and disadvantaged populations.”
In addition to the research initiative, the project will create new undergraduate and graduate curricula with modules, courses and certificates exploring the interplay of computer systems and AI at UW–Madison and the other institutions. The project’s initiatives for broadening participation are designed to cultivate leadership among underrepresented groups in AI and prepare them for AI and computer systems technology and research careers. These initiatives seek to benefit hundreds to thousands of students each year.
UW–Madison Computer Sciences department’s Michael Swift will co-lead the operating system design portion of this work along with Chris Rossbach at the University of Texas. Shivaram Venkataraman, also of UW–Madison Computer Sciences, will co-lead the effort on training ML models that can be deployed in operating systems with Aditya Akella at the University of Texas.
The grant is for $12 million over five years. Expeditions awards represent some of the largest investments provided by the NSF’s Directorate for Computer and Information Science and Engineering. The full name of the project is NSF Expeditions in Computing: Learning Directed Operating System (LDOS) – A Clean-Slate Paradigm for Operating Systems Design and Implementation.