The Python program dedupfs.py implements a file system in user space using FUSE. It's called DedupFS because the file system's primary feature is data deduplication, which enables it to store virtually unlimited copies of files because unchanged data is only stored once. In addition to deduplication the file system also supports transparent compression using the compression methods lzo, zlib and bz2. These properties make the file system ideal for backups: I'm currently storing 250 GB worth of backups using only 8 GB of disk space.
Several aspects of the design of DedupFS were inspired by Venti (ignoring the distributed aspect, for now…) and ZFS, though I've never personally used either. The ArchiveFS and lessfs projects share similar goals but have very different implementations.
The following shell commands show how to install and use the DedupFS file system on Ubuntu (where it was developed):
$ sudo apt-get install python-fuse $ git clone git://github.com/xolox/dedupfs.git $ mkdir mount_point $ python dedupfs/dedupfs.py mount_point # Now copy some files to mount_point/ and observe that the size of the two # databases doesn't grow much when you copy duplicate files again :-) # The two databases are by default stored in the following locations: # - ~/.dedupfs-metastore.sqlite3 contains the tree and meta data # - ~/.dedupfs-datastore.db contains the (compressed) data blocks
Development on DedupFS began as a proof of concept to find out how much disk space the author could free by employing deduplication to store his daily backups. Since then it's become more or less usable as a way to archive old backups, i.e. for secondary storage deduplication. It's not recommended to use the file system for primary storage though, simply because the file system is too slow. I also wouldn't recommend depending on DedupFS just yet, at least until a proper set of automated tests has been written and successfully run to prove the correctness of the code (the tests are being worked on).
The file system initially stored everything in a single SQLite database, but it turned out that after the database grew beyond 8 GB the write speed would drop from 8-12 MB/s to 2-3 MB/s. Therefor the file system now stores its data blocks in a separate database, which is a persistent key/value store managed by a dbm implementation like gdbm or Berkeley DB.
In the current implementation a file's content needs to fit in a cStringIO instance, which limits the maximum file size to your free RAM. Initially I implemented it this way because I was focusing on backups of web/mail servers, which don't contain files larger than 250 MB. Then I started copying virtual disk images and my file system blew up :-(. I know how to fix this but haven't implemented the change yet.
DedupFS was developed using Python 2.6, though it might also work on earlier versions. It definitely doesn't work with Python 3 yet though. It requires the Python FUSE binding in addition to several Python standard libraries like anydbm, sqlite3, hashlib and cStringIO.
If you have questions, bug reports, suggestions, etc. the author can be contacted at [email protected]. The latest version of DedupFS is available at http://peterodding.com/code/dedupfs/ and https://awesomeopensource.com/project/xolox/dedupfs.
This software is licensed under the MIT license.
© 2010 Peter Odding <[email protected]>.