On Wed, Jul 23, 2014 at 2:21 PM, Richard Pieri <[email protected]> wrote: > On 7/23/2014 2:02 PM, Bill Bogstad wrote: >> The problem seems to be 100% bad filesystem software. > > No, it's FUSE. > > FUSE runs in user space. Disk I/O happens in kernel space. This means > that read and write operations require much context switching. The > overhead for this is very high.
That certainly sounds plausible. But when I look for benchmarks for other FUSE based filesystems I see better numbers. It seems that other implementers using FUSE just do a better job the the NTFS-3g guys. Of course, the NTFS-3g guys also sell a commerical product so they have no incentive to improve the perrfomance of the free code. http://www.csl.sri.com/users/gehani/papers/SAC-2010.FUSE.pdf http://www.phoronix.com/scan.php?page=article&item=zfs_fuse_performance Now, I see some potential methodology issues with the Phoronix benchmark; but ZFS FUSE seems to do relatively well against the native filesystems with which it is being compared. Certainly much better then the FUSE based NTFS-3g results that I'm seeing for my rough and dirty testing. Bill Bogstad _______________________________________________ Discuss mailing list [email protected] http://lists.blu.org/mailman/listinfo/discuss
