junkurihara / rust-gd

An Implementation of Generalized Deduplication, written in Rust

rust-gd: A Rust Implementation of Generalized Deduplication

Rust implementation of Generalized Deduplication (GD) based on several types of error-correcting codes.

This is an implementation (and somewhat extension) of the novel concept of data deduplication method, called Generalized Deduplication (GD). The original concept of GD was introduced by a group of Aarhus University, Denmark, leaded by Prof. D. E. Lucani.

• Vestergaard, Rasmus, Qi Zhang, and Daniel E. Lucani. "Generalized deduplication: bounds, convergence, and asymptotic properties." 2019 IEEE Global Communications Conference (GLOBECOM). IEEE, 2019.
• Vestergaard, Rasmus, Daniel E. Lucani, and Qi Zhang. "Generalized deduplication: Lossless compression for large amounts of small IoT data." European Wireless 2019; 25th European Wireless Conference. VDE, 2019.
• etc.

Usage

Add the following to your Cargo.toml as imported directly from GitHub:

[dependencies]
rust-gd = { git = "https://github.com/junkurihara/rust-gd.git" }

or from crates.io:

[dependencies]
rust-gd = "*" // or appropriate version

Then, add use in your .rs file.

use rust_gd::*;

Example

NOTE: The compression rate strongly depends on the data alignment and data structure. So you should carefully choose the parameters according to the characteristics of given data.

GD with Reed-Solomon code over $\mathrm{GF}(2^8)$

use rust_gd::*;

let to_be_deduped: &[u8] =
    "寿限無(じゅげむ)寿限無(じゅげむ)五劫(ごこう)のすりきれ海砂利(かいじゃり)padpadpadpadpadpadpadpad".to_string().repeat(128).as_bytes()

let code_len = 128; // codeword length over GF(256), i.e., N in (N, K) RS code
let msg_len = 124;  // message length over GF(256), i.e., K in (N, K) RS code
let dict_size = 127; // max entry size of a dictionary used in GD process

// GD instance for deduplication (compress)
let mut gd_dedup = GD::ReedSolomon(code_len, msg_len).setup(dict_size).await.unwrap(); // Async API

// GD instance for duplication (decompress)
let mut gd_dup = GD::ReedSolomon(code_len, msg_len).setup(dict_size).await.unwrap(); // Async API

// struct Deduped = {pub data: Vec<u8>, pub last_chunk_pad_bytelen: usize}
let deduped: Deduped = gd_dedup.dedup(to_be_deduped).await.unwrap(); // Async API
println!("> Deduped data size is {} bytes", x.data.len());

let duped: Vec<u8> = gd_dup.dup(&deduped).await.unwrap(); // Async API.
println!("> Duped size {} bytes", y.len();

assert_eq!(duped, words);

In GD with RS codes, an approach of error-alignment can be employed by

// Linear transformation matrix used for error-alignment. This must be nonsinglar.
let trans: [&[u8; 4]; 4] = [
      &[1, 0, 0, 0],
      &[1, 1, 1, 4],
      &[1, 1, 3, 0],
      &[1, 2, 0, 0],
    ];

// Instantiation
let mut gd_dedup = GD::ReedSolomon(4, 3).setup(15).await.unwrap();
let mut gd_dup = GD::ReedSolomon(4, 3).setup(15).await.unwrap();

// Set error alignment
let res_dedup = gd_dedup.set_error_alignment(trans).await; // this simply returns Result<()>
let res_dup = gd_dup.set_error_alignment(trans).await;   // this simply returns Result<()>
assert!(res_dedup.is_ok());
assert!(res_dup.is_ok());

// then use gd instances to deduplicate/duplicate data as above.

For the detailed design of RS-code based implementation and the basic idea error-alignment, see DESIGN.md.

GD with Hamming code

let hamming_deg = 4;         // Degree m of (2^m - 1, 2^m - m -1) Hamming code
let hamming_dict_size = 511; // max entry size of a dictionary used in GD process

let to_be_deduped: &[u8] =
    "寿限無(じゅげむ)寿限無(じゅげむ)五劫(ごこう)のすりきれ海砂利(かいじゃり)padpadpadpadpadpadpadpad".to_string().repeat(128).as_bytes()

// GD instance for deduplication (compress)
let mut gd_dedup = GD::Hamming(hamming_deg).setup(hamming_dict_size).await.unwrap(); // Async API

// GD instance for duplication (decompress)
let mut gd_dup = GD::Hamming(hamming_deg).setup(hamming_dict_size).await.unwrap(); // Async API

// struct Deduped = {pub data: Vec<u8>, pub last_chunk_pad_bytelen: usize}
let deduped: Deduped = gd_dedup.dedup(to_be_deduped).await.unwrap(); Async API
println!("> Deduped data size is {} bytes", x.data.len());

let duped: Vec<u8> = gd_dup.dup(&deduped).await.unwrap(); // Async API.
println!("> Duped size {} bytes", y.len();

Codes in our implementation

Currently, our GD implementation is based only on Hamming and Reed-Solomon (RS) codes. The GD based on RS codes processes data chunks as byte stream. On the other hand, Hamming-based GD serves data chunks as bit stream.

For GD implementation using Hamming codes, Hamming code with the degree $m = 3$ of the code works in the internal libecc library of error-correcting codes, i.e., a case of the code length $n = 2^m - 1 = 7$. However, the Hamming code of $m = 3$ cannot be employed as the underlying linear code of Hamming-based GD. This is because the code length, i.e., $n=7$ bits, is not sufficient to deduplicate a "byte"-based data. In order to reasonably deduplicate byte-based data, byte alignment is needed. So, we omitted $m = 3$ and considers the parameter $m \geq 4$.

Byte alignment: Our implementation employs an encoding method that chunks message sequences in the unit of bytes. For example, if $(15, 11)$ Hamming code is employed, a 2-byte message is divided into two one byte (= 8 bits) sequences, and pads $15-8=7$ bits of zeros to each sequence to deal as a 15-bit codeword of Hamming code.

TODO

Following should be considered to be implemented.

• Benchmark for the performance of deduplication

• Optimization of math operations

• Deletion and deviation using PRNG (Yggdrasil paper)

• Golomb-Rice codes

Caveats

At this time this solution should be considered suitable for research and experimentation, further code and security review is needed before utilization in a production application.

License

Licensed under the MIT license, see LICENSE file.