Production FS: Adapt or Die
summary
Cloud Foundry teams evaluate actual root filesystem implementations: AUFS works but lacks mainline support and quotas; btrfs looks attractive for snapshotting and quotas but its GC spikes IOPS and destabilizes hosts; a loopback-sparse-file trick enforces quotas but hits a hard loop device limit 255 that caps containers per cell. The current engineering path wires garden → runC → a new rootfs tool that assembles image layers with overlay + XFS quotas, tested in CI and used as a stopgap while further filesystems are explored.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:48Cloud Foundry
- 2:06What Is a Filesystem
- 4:11What Is a Root Filesystem
- 10:51Demo of Pushing an Optical Foundry
- 15:22Why a Ufs
- 34:25Ubuntu Image
- 36:54Why Why Did We Choose Overlay and Xfs
related talks
29mCubeFS in Action: Empowering Users Through Case Studies
19mFlatFS: Flatten Hierarchical File System Namespace on Non-volatile Memories
56mThe evolution of storage on Linux
33mThe Evolution and Innovation of CubeFS
42mCan File Systems Survive in Data-centric World?
47mEnabling Large Block Size devices in Linux
44mWhat's New in Cloud Foundry Volume Services
01h 08mBtrfs – The Filesystem of the Future?