| duperemove(8) | System Manager's Manual | duperemove(8) |
duperemove - Find duplicate extents and print them to stdout
duperemove [options] files...
duperemove is a simple tool for finding duplicated extents and submitting them for deduplication. When given a list of files it will hash their contents on a block by block basis and compare those hashes to each other, finding and categorizing extents that match each other. When given the -d option, duperemove will submit those extents for deduplication using the Linux kernel extent-same ioctl.
duperemove can store the hashes it computes in a hashfile. If given an existing hashfile, duperemove will only compute hashes for those files which have changed since the last run. Thus you can run duperemove repeatedly on your data as it changes, without having to re-checksum unchanged data. For more on hashfiles see the --hashfile option below as well as the Examples section.
duperemove can also take input from the fdupes program, see the --fdupes option below.
Duperemove has two major modes of operation one of which is a subset of the other.
When run without -d (the default) duperemove will print out one or more tables of matching extents it has determined would be ideal candidates for deduplication. As a result, readonly mode is useful for seeing what duperemove might do when run with -d. The output could also be used by some other software to submit the extents for deduplication at a later time.
It is important to note that this mode will not print out all instances of matching extents, just those it would consider for deduplication.
Generally, duperemove does not concern itself with the underlying representation of the extents it processes. Some of them could be compressed, undergoing I/O, or even have already been deduplicated. In dedupe mode, the kernel handles those details and therefore we try not to replicate that work.
This functions similarly to readonly mode with the exception that the duplicated extents found in our "read, hash, and compare" step will actually be submitted for deduplication. An estimate of the total data deduplicated will be printed after the operation is complete. This estimate is calculated by comparing the total amount of shared bytes in each file before and after the dedupe.
files can refer to a list of regular files and directories or be a hyphen (-) to read them from standard input. If a directory is specified, all regular files within it will also be scanned. Duperemove can also be told to recursively scan directories with the '-r' switch.
If hashfile does not exist it will be created. If it exists, duperemove will check the file paths stored inside of it for changes. Files which have changed will be rescanned and their updated hashes will be written to the hashfile. Deleted files will be removed from the hashfile.
New files are only added to the hashfile if they are discoverable via the files argument. For that reason you probably want to provide the same files list and -r arguments on each run of duperemove. The file discovery algorithm is efficient and will only visit each file once, even if it is already in the hashfile.
Adding a new path to a hashfile is as simple as adding it to the files argument.
When deduping from a hashfile, duperemove will avoid deduping files which have not changed since the last dedupe.
Note: If you are piping filenames from another duperemove instance it is advisable to do so into a temporary file first as running duperemove simultaneously on the same hashfile may corrupt that hashfile.
Note: Hyperthreading can adversely affect performance of the extent finding stage. If duperemove detects an Intel CPU with hyperthreading it will use half the number of cores reported by the system for cpu bound tasks.
Unfortunately, some versions of Btrfs exhibit extremely poor performance in fiemap as the number of references on a file extent goes up. If you are experiencing the dedupe phase slowing down or 'locking up' this option may give you a significant amount of performance back.
Note: This does not turn off all usage of fiemap, to disable fiemap during the file scan stage, you will also want to use the --lookup-extents=no option.
Read hashes from a hashfile. A file list is not required with this option. Dedupe can be done if duperemove is run from the same base directory as is stored in the hash file (basically duperemove has to be able to find the files).
Write hashes to a hashfile. These can be read in at a later date and deduped from.
Dedupe the files in directory /foo, recurse into all subdirectories. You only want to use this for small data sets.
Use duperemove with fdupes to dedupe identical files below directory foo.
Duperemove can optionally store the hashes it calculates in a hashfile. Hashfiles have two primary advantages - memory usage and re-usability. When using a hashfile, duperemove will stream computed hashes to it, instead of main memory.
If Duperemove is run with an existing hashfile, it will only scan those files which have changed since the last time the hashfile was updated. The files argument controls which directories duperemove will scan for newly added files. In the simplest usage, you rerun duperemove with the same parameters and it will only scan changed or newly added files - see the first example below.
Dedupe the files in directory foo, storing hashes in foo.hash. We can run this command multiple times and duperemove will only checksum and dedupe changed or newly added files.
Don't scan for new files, only update changed or deleted files, then dedupe.
Add directory bar to our hashfile and discover any files that were recently added to foo.
List the files tracked by foo.hash.
Duperemove v0.11 is fast at reading and cataloging data. Dedupe runs will be memory limited unless the '--hashfile' option is used. '--hashfile' allows duperemove to temporarily store duplicated hashes to disk, thus removing the large memory overhead and allowing for a far larger amount of data to be scanned and deduped. Realistically though you will be limited by the speed of your disks and cpu. In those situations where resources are limited you may have success by breaking up the input data set into smaller pieces.
When using a hashfile, duperemove will only store duplicate hashes in memory. During normal operation then the hash tree will make up the largest portion of duperemove memory usage. As of Duperemove v0.11 hash entries are 88 bytes in size. If you know the number of duplicate blocks in your data set you can get a rough approximation of memory usage by multiplying with the hash entry size.
Actual performance numbers are dependent on hardware - up to date testing information is kept on the duperemove wiki (see below for the link).
Hashfiles are essentially sqlite3 database files with several tables, the largest of which are the files and hashes tables. Each hashes table entry is under 90 bytes though that may grow as features are added. The size of a files table entry depends on the file path but a good estimate is around 270 bytes per file.
If you know the total number of blocks and files in your data set then you can calculate the hashfile size as:
Hashfile Size = Num Hashes X 90 + Num Files X 270
Using a real world example of 1TB (8388608 128K blocks) of data over 1000 files:
8388608 * 90 + 270 * 1000 = 755244720 or about 720MB for 1TB spread over 1000 files.
Yes, Duperemove uses a transactional database engine and organizes db changes to take advantage of those features. The result is that you should be able to ctrl-c the program at any point and re-run without experiencing corruption of your hashfile.
Duperemove will print out an estimate of the saved space after a dedupe operation for you.
You can get a more accurate picture by running 'btrfs fi df' before and after each duperemove run.
Be careful about using the 'df' tool on btrfs - it is common for space reporting to be 'behind' while delayed updates get processed, so an immediate df after deduping might not show any savings.
At the moment duperemove can detect that some underlying extents are shared with other files, but it can not resolve which files those extents are shared with.
Imagine duperemove is examing a series of files and it notes a shared data region in one of them. That data could be shared with a file outside of the series. Since duperemove can't resolve that information it will account the shared data against our dedupe operation while in reality, the kernel might deduplicate it further for us.
This is a little complicated, but it comes down to a feature in Btrfs called _bookending_. The Btrfs wiki explains this in detail: http://en.wikipedia.org/wiki/Btrfs#Extents.
Essentially though, the underlying representation of an extent in Btrfs can not be split (with small exception). So sometimes we can end up in a situation where a file extent gets partially deduped (and the extents marked as shared) but the underlying extent item is not freed or truncated.
Yes. To be specific, duperemove does not deduplicate the data itself. It simply finds candidates for dedupe and submits them to the Linux kernel extent-same ioctl. In order to ensure data integrity, the kernel locks out other access to the file and does a byte-by-byte compare before proceeding with the dedupe.
Deduplication will lead to increased fragmentation. The blocksize chosen can have an effect on this. Larger blocksizes will fragment less but may not save you as much space. Conversely, smaller block sizes may save more space at the cost of increased fragmentation.
Deduplication is currently only supported by the btrfs and xfs filesystem.
The Duperemove project page can be found at https://github.com/markfasheh/duperemove
There is also a wiki at https://github.com/markfasheh/duperemove/wiki
| September 2016 | Version 0.11 |