We recently released a prototype of chunkfs using FUSE on Linux with basic functionality, and have begun work on a chunkfs fsck. Our preliminary performance measurements do not indicate any glaring performance problems as yet. The next step will be an implementation based on ext2. We are not using ext3 primarily because the main advantage of ext3 over ext2 is fast recovery after a crash; since the goal of chunkfs is extremely fast fsck, ext3's main advantage over ext2 is reduced (we only have to run fsck on the few chunks with metadata being actively modified at the time of the crash).
