1<p align="center"><img src="https://raw.githubusercontent.com/facebook/zstd/dev/doc/images/zstd_logo86.png" alt="Zstandard"></p> 2 3__Zstandard__, or `zstd` as short version, is a fast lossless compression algorithm, 4targeting real-time compression scenarios at zlib-level and better compression ratios. 5It's backed by a very fast entropy stage, provided by [Huff0 and FSE library](https://github.com/Cyan4973/FiniteStateEntropy). 6 7The project is provided as an open-source dual [BSD](LICENSE) and [GPLv2](COPYING) licensed **C** library, 8and a command line utility producing and decoding `.zst`, `.gz`, `.xz` and `.lz4` files. 9Should your project require another programming language, 10a list of known ports and bindings is provided on [Zstandard homepage](http://www.zstd.net/#other-languages). 11 12**Development branch status:** 13 14[![Build Status][travisDevBadge]][travisLink] 15[![Build status][AppveyorDevBadge]][AppveyorLink] 16[![Build status][CircleDevBadge]][CircleLink] 17[![Build status][CirrusDevBadge]][CirrusLink] 18 19[travisDevBadge]: https://travis-ci.org/facebook/zstd.svg?branch=dev "Continuous Integration test suite" 20[travisLink]: https://travis-ci.org/facebook/zstd 21[AppveyorDevBadge]: https://ci.appveyor.com/api/projects/status/xt38wbdxjk5mrbem/branch/dev?svg=true "Windows test suite" 22[AppveyorLink]: https://ci.appveyor.com/project/YannCollet/zstd-p0yf0 23[CircleDevBadge]: https://circleci.com/gh/facebook/zstd/tree/dev.svg?style=shield "Short test suite" 24[CircleLink]: https://circleci.com/gh/facebook/zstd 25[CirrusDevBadge]: https://api.cirrus-ci.com/github/facebook/zstd.svg?branch=dev 26[CirrusLink]: https://cirrus-ci.com/github/facebook/zstd 27 28## Benchmarks 29 30For reference, several fast compression algorithms were tested and compared 31on a server running Arch Linux (`Linux version 5.0.5-arch1-1`), 32with a Core i9-9900K CPU @ 5.0GHz, 33using [lzbench], an open-source in-memory benchmark by @inikep 34compiled with [gcc] 8.2.1, 35on the [Silesia compression corpus]. 36 37[lzbench]: https://github.com/inikep/lzbench 38[Silesia compression corpus]: http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia 39[gcc]: https://gcc.gnu.org/ 40 41| Compressor name | Ratio | Compression| Decompress.| 42| --------------- | ------| -----------| ---------- | 43| **zstd 1.4.0 -1** | 2.884 | 530 MB/s | 1360 MB/s | 44| zlib 1.2.11 -1 | 2.743 | 110 MB/s | 440 MB/s | 45| brotli 1.0.7 -0 | 2.701 | 430 MB/s | 470 MB/s | 46| quicklz 1.5.0 -1 | 2.238 | 600 MB/s | 800 MB/s | 47| lzo1x 2.09 -1 | 2.106 | 680 MB/s | 950 MB/s | 48| lz4 1.8.3 | 2.101 | 800 MB/s | 4220 MB/s | 49| snappy 1.1.4 | 2.073 | 580 MB/s | 2020 MB/s | 50| lzf 3.6 -1 | 2.077 | 440 MB/s | 930 MB/s | 51 52[zlib]: http://www.zlib.net/ 53[LZ4]: http://www.lz4.org/ 54 55Zstd can also offer stronger compression ratios at the cost of compression speed. 56Speed vs Compression trade-off is configurable by small increments. 57Decompression speed is preserved and remains roughly the same at all settings, 58a property shared by most LZ compression algorithms, such as [zlib] or lzma. 59 60The following tests were run 61on a server running Linux Debian (`Linux version 4.14.0-3-amd64`) 62with a Core i7-6700K CPU @ 4.0GHz, 63using [lzbench], an open-source in-memory benchmark by @inikep 64compiled with [gcc] 7.3.0, 65on the [Silesia compression corpus]. 66 67Compression Speed vs Ratio | Decompression Speed 68---------------------------|-------------------- 69 |  70 71A few other algorithms can produce higher compression ratios at slower speeds, falling outside of the graph. 72For a larger picture including slow modes, [click on this link](doc/images/DCspeed5.png). 73 74 75## The case for Small Data compression 76 77Previous charts provide results applicable to typical file and stream scenarios (several MB). Small data comes with different perspectives. 78 79The smaller the amount of data to compress, the more difficult it is to compress. This problem is common to all compression algorithms, and reason is, compression algorithms learn from past data how to compress future data. But at the beginning of a new data set, there is no "past" to build upon. 80 81To solve this situation, Zstd offers a __training mode__, which can be used to tune the algorithm for a selected type of data. 82Training Zstandard is achieved by providing it with a few samples (one file per sample). The result of this training is stored in a file called "dictionary", which must be loaded before compression and decompression. 83Using this dictionary, the compression ratio achievable on small data improves dramatically. 84 85The following example uses the `github-users` [sample set](https://github.com/facebook/zstd/releases/tag/v1.1.3), created from [github public API](https://developer.github.com/v3/users/#get-all-users). 86It consists of roughly 10K records weighing about 1KB each. 87 88Compression Ratio | Compression Speed | Decompression Speed 89------------------|-------------------|-------------------- 90 |  |  91 92 93These compression gains are achieved while simultaneously providing _faster_ compression and decompression speeds. 94 95Training works if there is some correlation in a family of small data samples. The more data-specific a dictionary is, the more efficient it is (there is no _universal dictionary_). 96Hence, deploying one dictionary per type of data will provide the greatest benefits. 97Dictionary gains are mostly effective in the first few KB. Then, the compression algorithm will gradually use previously decoded content to better compress the rest of the file. 98 99### Dictionary compression How To: 100 1011. Create the dictionary 102 103 `zstd --train FullPathToTrainingSet/* -o dictionaryName` 104 1052. Compress with dictionary 106 107 `zstd -D dictionaryName FILE` 108 1093. Decompress with dictionary 110 111 `zstd -D dictionaryName --decompress FILE.zst` 112 113 114## Build instructions 115 116### Makefile 117 118If your system is compatible with standard `make` (or `gmake`), 119invoking `make` in root directory will generate `zstd` cli in root directory. 120 121Other available options include: 122- `make install` : create and install zstd cli, library and man pages 123- `make check` : create and run `zstd`, tests its behavior on local platform 124 125### cmake 126 127A `cmake` project generator is provided within `build/cmake`. 128It can generate Makefiles or other build scripts 129to create `zstd` binary, and `libzstd` dynamic and static libraries. 130 131By default, `CMAKE_BUILD_TYPE` is set to `Release`. 132 133### Meson 134 135A Meson project is provided within [`build/meson`](build/meson). Follow 136build instructions in that directory. 137 138You can also take a look at [`.travis.yml`](.travis.yml) file for an 139example about how Meson is used to build this project. 140 141Note that default build type is **release**. 142 143### Visual Studio (Windows) 144 145Going into `build` directory, you will find additional possibilities: 146- Projects for Visual Studio 2005, 2008 and 2010. 147 + VS2010 project is compatible with VS2012, VS2013, VS2015 and VS2017. 148- Automated build scripts for Visual compiler by [@KrzysFR](https://github.com/KrzysFR), in `build/VS_scripts`, 149 which will build `zstd` cli and `libzstd` library without any need to open Visual Studio solution. 150 151### Buck 152 153You can build the zstd binary via buck by executing: `buck build programs:zstd` from the root of the repo. 154The output binary will be in `buck-out/gen/programs/`. 155 156## Status 157 158Zstandard is currently deployed within Facebook. It is used continuously to compress large amounts of data in multiple formats and use cases. 159Zstandard is considered safe for production environments. 160 161## License 162 163Zstandard is dual-licensed under [BSD](LICENSE) and [GPLv2](COPYING). 164 165## Contributing 166 167The "dev" branch is the one where all contributions are merged before reaching "master". 168If you plan to propose a patch, please commit into the "dev" branch, or its own feature branch. 169Direct commit to "master" are not permitted. 170For more information, please read [CONTRIBUTING](CONTRIBUTING.md). 171
