Tianyin Xu's NSF CAREER Award Presents an Innovative Avenue Focused on Configuration Management for Cloud Computing

1/24/2022 8:40:23 AM Aaron Seidlitz, Illinois CS

When it came time to write the proposal for his recently earned NSF CAREER Award, Illinois CS professor Tianyin Xu hesitated.

Headshot of professor Tianyin Xu, in a dark shirt with glasses standing in front of computer equipment.
Illinois CS professor Tianyin Xu credited faculty mentors for helping him learn to shape his research proposals. This, he said, led to his recent NSF CAREER Award.

“I knew I was supposed to submit a CAREER proposal. In fact, I was expected to submit it much earlier as the CAREER program is designed for early career development,” Xu said. “However, I kept procrastinating – I felt very nervous to think about a proposal that defines my early career. I found I was in a position of having many ideas but failed to connect the dots. So, I was thinking about waiting for another year.”

Fortunately, Professor Darko Marinov – a close mentor and collaborator of Xu – made the choice crystal clear. Xu, who focuses on research in Systems and Networking, was debating whether his ideas about configuration management for cloud computing and datacenter systems was worthy of the NSF CAREER Award. Marinov told him not only was it worthy, but that he should have no choice but to submit a proposal.

The message was clear; submit the proposal because it could very well be successful.

“I would not even have a page, if Darko did not push me. He convinced me that I should and was ready to do it,” Xu said. “The truth is that, three weeks before the due, I had only four bullets on a Google Doc. Darko knocked my office and asked for the draft to read. I was embarrassed.”

Marinov sat with Xu to outline what makes an effective proposal and opened Xu’s mind to how his research ideas are connected and can be coherently and systematically presented.

“Darko is always extremely helpful. It is not unfair to say that many of the ideas in the proposal are contributed by him,” Xu said. “One day, we discussed until 4 a.m., and I couldn’t fall asleep after the discussion – I was too excited. Since then, I became very productive and wrote up the entire proposal in less than three weeks.”

Xu also learned how to write a good grant proposal from another faculty mentor, Professor Josep Torrellas.

“I was able to write a proposal in a very short amount of time, because I learned proposal writing from Josep,” Xu said. “I had multiple rejections in my first year at Illinois CS, but that began to change when I worked with Josep on a collaborative proposal. He taught me how to write a good proposal. After that, I became very successful in proposal writing and fundraising.

“I am always deeply grateful to Darko and Josep – I have the best mentors in the world.”

Additionally, Xu felt thankful for Illinois CS, especially Tandy Warnow, Grainger Distinguished Chair in Engineering and Associate Head for Research and Faculty Development, and Department Head Nancy M. Amato, for building a strong faculty mentorship program and for organizing many events to help junior faculty with fundraising.

The CAREER award itself covers more than $500,000 over the next five years, so Xu can further research on a topic titled, “Rethinking Configuration Management for Cloud and Datacenter Systems.”

Dating back to his time as a PhD student at University of California San Diego, Xu has remained interested in the impact configuration management can have on cloud and datacenter computing.

“It is still a largely unsolved problem. Bad configuration changes are still the major cause of catastrophic system failures and service outages, and there is, unfortunately, no effective solution,” Xu said.

His early research and industry experience lead to his vision of continuous configuration testing to improve the reliability of cloud-scale systems.

“My research interest is in the reliability of cloud-scale systems,” Xu said. “My methodology is to understand why existing computer systems fail to deliver the desired reliability, despite the decades of research and engineering efforts, and then to think through the fundamental opportunities for innovation and technical contributions. Configuration research is exactly done this way.”

Having worked at Facebook’s Core Systems as a visiting scientist prior to joining Illinois CS, Xu was well equipped with the understanding of these needs and challenges.

He joined Facebook based on his PhD research on misconfiguration detection. Citing the “impact-driven nature” of working in industry, Xu said he did not immediately start to productionize his PhD research, but took the time to understand the potential impacts of his early work and the other related research ideas.

After a careful analysis of service-level disruptions of Facebook systems, Xu didn’t gather as many of the benefits as he was hoping for.

“I realized that much of the research at the time, including my own work, addressed a limited number of real-world configuration-induced failures,” Xu said. “While it is still worthwhile to further improve the work given the high costs and severe consequences of those failures, I was instead much more intrigued by the large numbers of incidents that were fundamentally beyond the capability of the research then.”

“And that became the next round of research I have conducted at Illinois CS.”

His work on configuration testing has already yielded significant results, award-winning papers, and became the basis of this NSF CAREER Award.

Going back to the role Marinov plays in this project’s development, Xu also said the active interaction between the two became critical to the research progress.

“One question Darko always asks is simple, but instrumental: ‘Why can't your technique be deployed tomorrow at Facebook?’” Xu said. “It drives me crazy sometimes, but it pushes me to think deeper and inspires follow-up research.”

With that in mind, his NSF CAREER work aims to, according to his abstract, “(1) develop novel, practical technologies to address key problems of existing configuration management techniques, and (2) explore radically new system designs to prevent configuration problems in the first place.”

Additionally, Xu is proud that the work is interdisciplinary and will include undergraduate students with diverse backgrounds. Considering his devotion to the department’s Broadening Participation in Computing initiative, Xu believes this work will be a great opportunity to engage junior students.

“Certainly, there are a lot of hard problems we need to address, but I believe that the research will lead to practical, impactful solutions to improve cloud reliability,” Xu said. “We are incredibly excited and more ready than ever to drive the research in the next five years.”