amoghe / dedup

Deduplicator

dedup - a deduplication tool (+ library)

Why

Deduplication can be thought of as a coarse grained compression that detects duplicate data over much larger windows than most compressors work across. As a result, deduplication is a good first step and passing deduplicated data into a downstream compressor often results in much better compression performance (in terms of compression ratio and sometimes speed)

Library

dedup is a golang lib that allows for arbitrary data to be deduplicated (input from an io.Reader). The dedup.Deduplicator and dedup.Reduplicator are the workhorses that actually do all the work. For examples on how to use them, see the dedup tool itself (in cmd/dedup/main.go).

Installation

As with any go(lang) pkg, you can go get by

shell> go get github.com/amoghe/dedup

Usage

import (
  "github.com/amoghe/dedup"
)

// somewhere in your code
err := dedup.NewDeduplicator(windowSize, mask).Do(os.Stdin, os.Stdout)

err := dedup.NewReduplicator().Do(os.Stdin, os.Stdout)

Binary

This codebase also builds a cmdline tool named dedup (see cmd/dedup) that can be used to deduplicate data.

Installation

The tool can be installed by either building from source

shell> go get github.com/amoghe/dedup && \
  cd $GOPATH/src/github.com/amoghe/dedup && \
  go install

Alternatively, you can download a release binary from the Releases section of this github project.

Usage

Consider this workload where we save two similar docker containers:

akshay@spitfire:~/$ time docker save redmine bitnami/redmine | gzip | wc --bytes
497548816 # <-- 474.49 MB (or MiB)

real  1m7.900s
user  0m58.536s
sys   0m1.780s

akshay@spitfire:~/$ time docker save redmine bitnami/redmine | dedup | gzip | wc --bytes
295793793 # <-- 282.09 MB (or MiB)

real  0m50.261s
user  0m56.312s
sys   0m3.688s

As you can see, some workloads can benefit greatly from a combination of deduplication + compression (in terms of both compression ratio and speed)

Compression

Note that this lib (and tool) probably won't ever support built-in support for compression of the output stream. You should pick an appropriate compressor "downstream" from this lib/tool. You'll find that standalone compressors such as gzip, bzip2, xz (and their parallel implementations - pigz, pbzip2, pxz) are readily available on most linux distributions. These compressors support pipelining (i.e. i/o can be pipelined via the shell) so there is no need for this library to provide this functionality.

TODO:

• Currently the deduplication lib consumes memory that is proportional to the size of the input file. (See issue #1)
• Document the usage and impact of the windowSize and zeroBits parameters used by the Deduplicator
• Add progress reporting when input is a large file (not stdin)
• Make cmdline args fully compatible with other compression tools ('-k', '-v')
• Add tests! (unit tests, fuzz tests)

LICENSE

See the LICENSE file