Rethinking Writeback - Scaling Linux Filesystem and Memory Performance for the Next Decade

Explore a comprehensive technical conference talk examining the evolution of Linux's writeback infrastructure to meet modern computing demands. Discover how current writeback mechanisms, originally designed before large folios, CXL-tiered memory, and AI workloads, are reaching scalability limits with contemporary applications like RAG pipelines using vector databases and memory tiering systems. Learn about the challenges posed by single-threaded design that stalls page migration, reduces memory compaction effectiveness, and impacts hugepage allocations and folio movement across memory tiers. Understand how these limitations contribute to memory fragmentation and affect performance in sustained-write workloads with large memory footprints on high-capacity SSDs. Examine proposed solutions for parallelizing writeback to improve throughput and responsiveness under dirty-page pressure, and explore early kernel community experiments focused on making writeback more filesystem-geometry aware and parallelizing based on overwrites versus new allocations. Gain insights into reimagining writeback as a scalable, performance-critical component essential for Linux's future in handling next-generation workloads and memory architectures.