No, FUSE is really slow, it just depends what you're benchmarking. That guy is really benchmarking the disk, not the filesystem. If you benchmark things that

No, FUSE is really slow, it just depends what you're benchmarking. That guy is really benchmarking the disk, not the filesystem. If you benchmark things that test the filesystem layer on its operations---that is, make it do lots of things to metadata but not read or write much actual data---FUSE gets the crap beaten out of it. As an example, write a no-op filesystem in FUSE and a no-op filesystem under the VFS, and run postmark[1] on it. You should get some pretty drastic results. My professor has something called FiST[2] (like yacc for filesystem devs) that gets about 1-2% overhead over a native FS, and I believe one of his papers contains a comparison with FUSE that shows how bad FUSE is, but I don't remember it. [1]: http://www.shub-internet.org/brad/FreeBSD/postmark.html http://www.shub-internet.org/brad/FreeBSD/postmark.html [2]: http://www.fsl.cs.sunysb.edu/project-fist.html http://www.fsl.cs.sunysb.edu/project-fist.html