We’ve all experienced an untimely computer crash while working on an important file or playing a game, only to reboot and discover that the document or game progress has been corrupted—or worse, lost. As a step toward a potential fix, Department of Computer Sciences (CS) alumna Jing Liu PhD’24, graduate student Yifan Dai PhDx’25, and Professors Andrea Arpaci-Dusseau and Remzi Arpaci-Dusseau built a system to allow computers to recover from this type of crash quickly and smoothly without needing to fully reboot.
Their system, called Ananke, won the Best Paper Award at the 2025 USENIX Conference on File and Storage Systems Technologies. Liu goes on to say that one advantage of Ananke is that “your application can keep running smoothly as if nothing happened, making data loss and downtime much less likely.”
The paper, titled “Fast, Transparent Filesystem Microkernel Recovery with Ananke,” outlines “a high-performance filesystem microkernel service designed to provide fast and transparent recovery from filesystem crashes.” The paper discusses “how Ananke leverages microkernel architectures to enable process-level recovery, thereby maintaining application continuity without requiring full-system reboots,” explains Liu. “Ananke introduces key techniques such as the P-Crash Log (P-Log), the Act-Ignore-Modify (AIM) algorithm, kernel-coordinated speculative restart, and lightweight error detection with checksums.”
Liu came to UW-Madison to study with Andrea and Remzi Arpaci-Dusseau, whom she first came in contact with when reading the Wikipedia page for the Linux ext4 filesystem, which referenced an article written by the Arpaci-Dusseaus and others. The Wikipedia page describes the idea of journal checksumming, which inspired a key feature of ext4’s reliability. “Given that ext4 is the default filesystem in most modern Linux desktops and cloud environments, this makes it a widely used innovation. It’s pretty difficult for a research idea to become a key feature in the widely used modern operating systems,” Liu says. “I think having that level of impact—where your research becomes part of everyone’s daily life—is truly meaningful.”
As a senior researcher at Microsoft Research-Asia, Liu plans to build advanced systems that support the rising wave of next-generation computational infrastructure and highly autonomous systems. “By improving resource efficiency and enhancing user experience, I aim to pave the way for more efficient and ubiquitous AI-driven technologies,” says Liu. As AI techniques become deeply integrated into everyday life, making it crucial to build efficient systems, Liu hopes “future research will help lay the foundation for next-generation systems.”
The paper was presented at FAST ’25, the 23rd USENIX Conference on File and Storage Technologies, hosted in Santa Clara, CA at the end of February, 2025.