Skip to content
This repository has been archived by the owner on Sep 21, 2023. It is now read-only.
/ fastcdc-go Public archive
forked from jotfs/fastcdc-go

Go implementation of the FastCDC content-defined chunking algorithm

License

Notifications You must be signed in to change notification settings

kastelo-labs/fastcdc-go

 
 

Repository files navigation

FastCDC-Go

Docs Build Status codecov Go Report Card

FastCDC-Go is a Go library implementing the FastCDC content-defined chunking algorithm.

Install:

go get -u github.com/jotfs/fastcdc-go

Example

import (
  "bytes"
  "fmt"
  "log"
  "math/rand"
  "io"

  "github.com/jotfs/fastcdc-go"
)

opts := fastcdc.Options{
  MinSize:     256 * 1024
  AverageSize: 1 * 1024 * 1024
  MaxSize:     4 * 1024 * 1024
}

data := make([]byte, 10 * 1024 * 1024)
rand.Read(data)
chunker, _ := fastcdc.NewChunker(bytes.NewReader(data), opts)

for {
  chunk, err := chunker.Next()
  if err == io.EOF {
    break
  }
  if err != nil {
    log.Fatal(err)
  }

  fmt.Printf("%x  %d\n", chunk.Data[:10], chunk.Length)
}

Command line tool

This package also includes a useful CLI for testing the chunking output. Install it by running:

go install ./cmd/fastcdc

Example:

# Outputs the position and size of each chunk to stdout 
fastcdc -csv -file random.txt

Performance

FastCDC-Go is fast. Chunking speed on an Intel i5 7200U is >1GiB/s. Compared to restic/chunker, another CDC library for Go, it's about 2.9 times faster.

Benchmark (code):

BenchmarkRestic-4     23384482467 ns/op	 448.41 MB/s	 8943320 B/op   15 allocs/op
BenchmarkFastCDC-4    8080957045 ns/op	1297.59 MB/s	16777336 B/op    3 allocs/op

Normalization

A key feature of FastCDC is chunk size normalization. Normalization helps to improve the distribution of chunk sizes, increasing the number of chunks close to the target average size and reducing the number of chunks clipped by the maximum chunk size, as compared to the Rabin-based chunking algorithm used in restic/chunker.

The histograms below show the chunk size distribution for fastcdc-go and restic/chunker on 1GiB of random data, each with average chunk size 1MiB, minimum chunk size 256 KiB and maximum chunk size 4MiB. The normalization level for fastcdc-go is set to 2.

Compared the restic/chunker, the distribution of fastcdc-go is less skewed (standard deviation 345KiB vs. 964KiB).

License

FastCDC-Go is licensed unser the Apache 2.0 License. See LICENSE for details.

References

  • Xia, Wen, et al. "Fastcdc: a fast and efficient content-defined chunking approach for data deduplication." 2016 USENIX Annual Technical Conference pdf

About

Go implementation of the FastCDC content-defined chunking algorithm

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Go 100.0%