ScaleXFS - Getting Scalability of XFS Back on the Ring
Explore the scalability challenges of XFS journaling in manycore systems through this 14-minute conference talk from FAST '22. Delve into the primary causes of XFS scalability failure, including contention between in-memory and on-disk logging, as well as among multiple concurrent in-memory loggings. Learn about three key techniques proposed to address these issues: Double Committed Item List, Per-core In-memory Logging, and Strided Space Counting. Discover how these innovations, collectively known as ScaleXFS, significantly improve performance, reducing metadata operation latency and increasing throughput in various workloads. Gain insights into hardware trends, journaling in filesystems, and the scalable design of Ext4 for comparison. Examine evaluation results demonstrating ScaleXFS's superior performance over stock XFS in terms of throughput and latency reduction.