Ruidan Li, University of Chicago; Erci Xu, Shanghai Jiao Tong University; Xikang Song, University of Chicago; Kiran Srinivasan, NetApp; Haryadi S. Gunawi, University of Chicago
This paper presents a comprehensive study of fail-slow behaviors using the largest dataset available to date. We identified five fail-slow patterns, four of which were previously unexamined, and found that all state-of-the-art approaches fail to capture them fully. Hence, we build KRATOS, a comprehensive, robust, and practical detection method that accurately identifies all patterns, remains effective across diverse environments without fine-tuning, and operates efficiently at scale. On multiple RAID and non-RAID datasets (including SSDs and HDDs), KRATOS detects 32.8-50% more cases, achieves over 95% accuracy across a broad threshold range, and runs 3-128× faster than existing methods.
Open Access Media
USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.
