Viewing: TODO
BACKGROUND
----------
Implement an MemFS-based OSD device with writeback support for Lustre.
It borrows lots of design from memory-based file systems such as tmpfs/ramfs.
The data is frist written into the memory-based file system (called MemFS in
short). And then, the data can be persisted to the permant storage in a delayed
writeback manner.
+---------------------------------------------------------+
| This is experimental! Do NOT use for important data! |
| Only bugs and data corruption lie ahead! Turn back now! |
+---------------------------------------------------------+
For questions, please contact:
- Yingjin Qian <qian@ddn.com>
- Timothy Day <timday@amazon.com>
TODO
----
- Inode and space usage accounting for statfs() system call.
- Limiting for inodes and blocks.
- Refine the mount command support for MemFS-based OSD.
- lprocfs support. Track OSD stats and access them via lprocfs.
- Use Maple Tree in new kernel to manage and access entries within a directory.
- Implement the functionality needed by LFSCK.
- Quota support.
- Swap space support for large files.
- Metadata on MemFS; Data on Persistent storage
(just like PCC naming with FID for data).
- Writeback support with ldiskfs/ZFS or KV store as persistent backends.
- Add transcation support.