dwarfs

DwarFS

The Deduplicating Warp-speed Advanced Read-only File System.

A fast high-compression read-only file system for Linux, FreeBSD, macOS and Windows.

Table of contents

• What is DwarFS (in plain words)?
• Why not just use .zip or .tar.gz?
• Performance comparison overview
• Quick Start
• Overview
• History
• Building and Installing
  • Note to Package Maintainers
  • Prebuilt Binaries
  • Universal Binaries
  • Dependencies
  • Building
  • Installing
  • Static Builds
• Usage
• Using the Libraries
• Windows Support
  • Building on Windows
• macOS Support
• Use Cases
  • Astrophotography
• Dealing with Bit Rot
• Extended Attributes
• Comparison
  • With SquashFS
  • With SquashFS & xz
  • With lrzip
  • With zpaq
  • With zpaqfranz
  • With wimlib
  • With Cromfs
  • With EROFS
  • With fuse-archive
• Performance Monitoring
• Other Obscure Features
• Related projects
• Notable users
• Stargazers over Time

What is DwarFS (in plain words)?

DwarFS is a mountable archive: you pack a directory into one file (an “image”), then open it instantly like a normal folder — no full extraction, no temporary files. It’s read-only when mounted, so you can browse, open, and run files safely in place. This works particularly well for big collections with lots of similar or repeated content — think many versions of a project or dataset, folders of documents and text files, backups/snapshots that mostly overlap, or libraries with many near-duplicates. The image can be mounted in fractions of a second, you can use the contents immediately with no delay, and you extract only if you actually need to.

Why not just use .zip or .tar.gz?

Traditional archives are fine for storage, but they’re usually slow to open and awkward for random access (jumping around inside large files or across many files). DwarFS is built for fast random reads and space savings. It groups similar files and removes duplication, so images are often smaller than a simple tar/zip, while reads stay snappy — even when accessing lots of files simultaneously. In practice, you keep huge directories compressed, mount them in milliseconds, and work as if they were already unpacked.

Performance comparison overview

This comparison uses DwarFS 0.14.1, 7-Zip 25.00 (x64), and tar + pigz for all .tar.gz tests since plain gzip would have been incredibly slow. Both .tar.gz and 7zip archives were mounted using fuse-archive v1.16 with the -olazycache option since the default of caching the entire archive in memory is infeasible for large archives.

1139 complete Perl installations

Perl (47.49 GiB, 1.9M files)	.tar.gz [^pl1]	.tar.zst [^pl2]	7zip (-mx=7)	SquashFS [^pl7]	DwarFS (lzma)	DwarFS (zstd)
Compression time	4m 59s	8m 06s	23m 27s	5m 37s	2m 13s	5m 3s
Compression CPU time	1h 47m	2h 18m	5h 5m	2h 29m	31m 17s	49m 51s
Compressed size	12.17 GiB	0.387 GiB	1.219 GiB	3.245 GiB	0.310 GiB	0.352 GiB
Compression ratio	3.902	122.7	38.96	14.63	153.2	134.9
Decompression time	2m 19s	57.2s	1m 14s	39.3s	48.7s	46.5s
Decompression CPU time	3m 44s	1m 21s	2m 28s	1m 25s	2m 23s	1m 59s
Mount time	2m 07s	❌ [^pl3]	3.638s	0.011s	0.420s	0.009s
Find all 1.9M files [^pl8]	5.670s	❌ [^pl3]	5.695s	5.311s	2.800s	2.821s
Checksum 1139 files (2.58 GiB) [^pl4]	❌ [^pl5]	❌ [^pl3]	~5h [^pl6]	1.541s	4.330s	1.134s

[^pl1]: using pigz -9 [^pl2]: using zstd --long=31 --ultra -22 -T0 [^pl3]: not supported by fuse-archive [^pl4]: $ ls -1 mnt/*/perl*/bin/perl5* | xargs -d $'\n' -n1 -P16 sha256sum [^pl5]: killed after making no progress for 15 minutes [^pl6]: killed when only 78 files were finished after about 20 minutes [^pl7]: using -comp zstd -Xcompression-level 22 -b 1M -tailends; using squashfuse_ll 0.6.0 [^pl8]: $ fd -t f . mnt | wc -l

All artifacts from 205 DwarFS CI builds

DwarFS CI (465.2 GiB, 3.6M files)	.tar.gz (pigz -9)	7zip (-mx=7)	DwarFS (zstd)
Compression time	40m 28s	141m 46s	68m 6s
Compression CPU time	13h 55m	42h 47m	31h 12m
Compressed size	142.9 GiB	60.04 GiB	28.63 GiB
Compression ratio	3.255	7.748	16.25
Decompression time	26m 17s	8m 15s	7m 21s
Decompression CPU time	35m 16s	41m 29s	8m 59s
Mount time	22m 18s	10.5s	0.024s
Run 441 binaries (2.72 GiB) [^ci1]	❌ [^ci2]	❌ [^ci3]	0.774s

[^ci1]: $ ls -1 mnt/*/dwarfs-*-Linux-x86_64-*-minsize*/bin/mkdwarfs | xargs -d $'\n' -n1 -P16 sh -c '$0 -h' [^ci2]: killed after making no progress for 15 minutes [^ci3]: killed after making no progress and consuming more than 32 GiB of RAM

Game audio assets from sonniss.com

Sonniss (3.072 GiB, 171 files) ^wav1	.tar.gz (pigz -9)	7zip (-mx=7)	SquashFS [^wav3]	DwarFS (categorize)
Compression time	5.34s	6m 21s	19.52s	3.98s
Compression CPU time	2m 1s	19m 14s	9m 47s	29.0s
Compressed size	2.725 GiB	2.255 GiB	2.711 GiB	1.664 GiB
Compression ratio	1.127	1.362	1.133	1.846
Decompression time	13.7s	1m 50s	0.774s	1.32s
Decompression CPU time	17.8s	1m 52s	4.60s	9.15s
Mount time	13.6s	0.014s	0.018s	0.008s
Checksum all files [^wav2]	3m 2s	30m 42s	2.81s	2.64s

[^wav2]: $ find mnt -type f | xargs -d $'\n' -n1 -P16 sha256sum [^wav3]: using -comp zstd -Xcompression-level 22 -b 1M -tailends

Quick Start

To create a DwarFS image from a directory, use mkdwarfs:

$ mkdwarfs -i /path/to/input/dir -o /path/to/image.dwarfs

To mount the image, use the dwarfs FUSE driver:

$ mkdir /path/to/mountpoint
$ dwarfs /path/to/image.dwarfs /path/to/mountpoint
$ ls /path/to/mountpoint

To extract the image, use dwarfsextract:

$ dwarfsextract -i /path/to/image.dwarfs -o /path/to/output/dir

There’s also the dwarfsck tool that can be used for a variety of tasks, for example listing the files in a DwarFS image:

$ dwarfsck /path/to/image.dwarfs -l

It can also be used for generating checksums for each file in the image in a format that is recognized by the *sum commands:

$ dwarfsck /path/to/image.dwarfs --checksum=sha512 | sha512sum --check

This is useful if you want to verify the integrity of the actual files stored in the DwarFS image.

Overview

DwarFS is a read-only file system with a focus on achieving very high compression ratios, particularly for highly redundant data.

This probably doesn’t sound very exciting, because if it’s redundant, it should compress well. However, I found that other read-only, compressed file systems don’t do a very good job at making use of this redundancy. See here for a comparison with other compressed file systems.

DwarFS also doesn’t compromise on speed; in my use cases, it performs on par with, or better than, SquashFS. For my primary use case, DwarFS compression is an order of magnitude better than SquashFS compression, it’s 6 times faster to build the file system, it’s typically faster to access files on DwarFS and it uses less CPU resources.

To give you an idea of what DwarFS is capable of, here’s a quick comparison of DwarFS and SquashFS on a set of video files with a total size of 39 GiB. The twist is that each unique video file has two sibling files with a different set of audio streams (this is an actual use case). So there’s redundancy in both the video and audio data, but as the streams are interleaved and identical blocks are typically very far apart, it’s challenging to make use of that redundancy for compression. SquashFS essentially fails to compress the source data at all, whereas DwarFS is able to reduce the size to nearly one-third, which is close to the theoretical maximum:

$ du -hs dwarfs-video-test
39G     dwarfs-video-test
$ ls -lh dwarfs-video-test.*fs
-rw-r--r-- 1 mhx users 14G Jul  2 13:01 dwarfs-video-test.dwarfs
-rw-r--r-- 1 mhx users 39G Jul 12 09:41 dwarfs-video-test.squashfs

Furthermore, when mounting the SquashFS image and performing a random-read throughput test using fio-3.34, both squashfuse and squashfuse_ll top out at around 230 MiB/s:

$ fio --readonly --rw=randread --name=randread --bs=64k --direct=1 \
      --opendir=mnt --numjobs=4 --ioengine=libaio --iodepth=32 \
      --group_reporting --runtime=60 --time_based
[...]
   READ: bw=230MiB/s (241MB/s), 230MiB/s-230MiB/s (241MB/s-241MB/s), io=13.5GiB (14.5GB), run=60004-60004msec

In comparison, DwarFS manages to sustain random read rates of 20 GiB/s:

  READ: bw=20.2GiB/s (21.7GB/s), 20.2GiB/s-20.2GiB/s (21.7GB/s-21.7GB/s), io=1212GiB (1301GB), run=60001-60001msec

Distinct features of DwarFS are:

• Clustering of files by similarity using a similarity hash function. This makes it easier to exploit the redundancy across file boundaries.

• Segmentation analysis across file system blocks in order to reduce the size of the uncompressed file system. This saves memory when using the compressed file system and thus potentially allows for higher cache hit rates as more data can be kept in the cache.

• Categorization framework to categorize files or even fragments of files and then process individual categories differently. For example, this allows you to not waste time trying to compress incompressible files or to compress PCM audio data using FLAC compression.

• Highly multi-threaded implementation. Both the file system creation tool as well as the FUSE driver are able to make good use of the many cores of your system.

History

I started working on DwarFS in 2013 and my main use case and major motivation was that I had several hundred different versions of Perl that were taking up something around 30 gigabytes of disk space, and I was unwilling to spend more than 10% of my hard drive keeping them around for when I happened to need them.

Up until then, I had been using Cromfs for squeezing them into a manageable size. However, I was getting more and more annoyed by the time it took to build the file system image and, to make things worse, more often than not it was crashing after about an hour or so.

I had obviously also looked into SquashFS, but never got anywhere close to the compression rates of Cromfs.

This alone wouldn’t have been enough to get me into writing DwarFS, but at around the same time, I was pretty obsessed with the recent developments and features of newer C++ standards and really wanted a C++ hobby project to work on. Also, I’ve wanted to do something with FUSE for quite some time. Last but not least, I had been thinking about the problem of compressed file systems for a bit and had some ideas that I definitely wanted to try.

A lot of the initial code was written in 2013, then I did a couple of cleanups, bugfixes and refactors every once in a while, but I never really got it to a state where I would feel happy releasing it. It was too awkward to build with its dependency on Facebook’s folly library and it didn’t have any documentation.

Digging out the project again in 2020, things didn’t look as grim as they used to. Folly was building with CMake and so I just pulled it in as a submodule. Most other dependencies could be satisfied from packages that should be widely available.

However, the folly and fbthrift dependencies were still a bit of a pain. DwarFS used only a small fraction of folly, and an even tinier fraction of fbthrift. Folly had so many dependencies that it was getting harder and harder to build. Often, when upgrading, there were new build problems. So in 2026, after some of my pull requests for folly and fbthrift had been stuck in review with no feedback for more than half a year, I decided to ditch folly, fork the bits I needed from fbthrift, and re-write the thrift code generation in Python. That was quite a bit of work, but less than I anticipated. A lot of the folly features could be replaced by Boost or the C++ standard library. Some other, more unique features had to be re-implemented. This, however, gave me the freedom to just fix the things I previously had pull requests for, and it turned out that the new implementation was measurably faster in some cases. Also, some re-implementations were kept much simpler than the original folly code, making the resulting binaries smaller. A bunch of dependencies, such as gflags, glog, libevent, double-conversion, and fast_float, could be removed entirely.

Building and Installing

Note to Package Maintainers

DwarFS should usually build fine with minimal changes out of the box. If it doesn’t, please file an issue. I’ve set up CI jobs using Docker images for Ubuntu (22.04 and 24.04), Fedora Rawhide, openSUSE Tumbleweed, Arch Linux, and Debian, as well as a setup script for FreeBSD, that can help with determining an up-to-date set of dependencies. Note that building from the release tarball requires less dependencies than building from the git repository, notably the ronn tool as well as Python and the mistletoe Python module are not required when building from the release tarball.

There are some things to be aware of:

• While supported by the CMakeLists.txt via -DPREFER_SYSTEM_GTEST=ON, it is not recommended to use a system-installed version of GoogleTest. GoogleTest itself recommends downloading it as part of the build, and issues like this one can arise quickly if the system-installed version is built with different compiler flags than the DwarFS build.

• For other bundled libraries (namely fmt, parallel-hashmap, range-v3), the system-installed version is used as long as it meets the minimum required version. Otherwise, the preferred version is fetched during the build.

Prebuilt Binaries

Each release has pre-built, statically linked binaries for a number of architectures available for download. These should run without any dependencies and can be useful especially on older distributions where you can’t easily build the tools from source. For example, these binaries will work perfectly fine on Linux distributions from 2007.

Universal Binaries

In addition to the binary tarballs, there’s a universal binary available for each architecture. These universal binaries contain all tools (mkdwarfs, dwarfsck, dwarfsextract and the dwarfs fuse3 driver) in a single executable. These executables are self-extracting and use either a custom self-extractor or upx. They are generally optimized to provide the best performance in the smallest possible binary and are much smaller than the individual tools combined. However, this also means the binaries need to be decompressed each time they are run, which can add significant overhead. If that is an issue, you can either stick to the “classic” individual binaries or you can decompress the universal binary. For upx-compressed binaries, you can use:

$ upx -d dwarfs-universal-0.7.0-Linux-aarch64

For the binaries that use the custom self-extractor, you can use:

$ ./dwarfs-universal-riscv64 --extract-wrapped-binary dwarfs-universal

Note that both self-extractors need at least Linux kernel 3.17 to work properly. If you want to use the FUSE driver, you’ll need to install the fuse3 tools for your distribution. If you want to run the binaries on an older kernel, you can unpack the universal binary (unpacking does not require kernel 3.17). If you’re stuck with fuse2, you must use the individual dwarfs2 driver instead of the universal binary.

You can run the universal binaries via symbolic links named after the tool. For example:

$ ln -s dwarfs-universal-0.7.0-Linux-aarch64 mkdwarfs
$ ./mkdwarfs --help

This also works on Windows if the file system supports symbolic links:

> mklink mkdwarfs.exe dwarfs-universal-0.7.0-Windows-AMD64.exe
> .\mkdwarfs.exe --help

Alternatively, you can select the tool by passing --tool=<name> as the first argument on the command line:

> .\dwarfs-universal-0.7.0-Windows-AMD64.exe --tool=mkdwarfs --help

Note that just like the dwarfs.exe Windows binary, the universal Windows binary depends on the winfsp-x64.dll from the WinFsp project. However, for the universal binary, the DLL is loaded lazily, so you can still use all other tools without the DLL. See the Windows Support section for more details.

Dependencies

DwarFS uses CMake as a build tool.

If you build from a release tarball, the build will be faster and there will be fewer dependencies, because the release tarball ships with the auto-generated files that would otherwise have to be generated during the in-repository build.

You can grab an up-to-date list of dependencies from the docker files used for the CI builds:

• Arch Linux
• Debian Testing
• Fedora Rawhide
• openSUSE Tumbleweed
• Ubuntu 22.04
• Ubuntu 24.04

For FreeBSD, you can use PKGS from the script that sets up the jail for the CI builds.

Building

First, unpack the release tarball:

$ tar xvf dwarfs-x.y.z.tar.xz
$ cd dwarfs-x.y.z

Alternatively, you can also clone the git repository, but be aware that this has more dependencies and the build will likely take longer because the release archive ships with most of the auto-generated files that will have to be generated when building from the repository:

$ git clone --recurse-submodules https://github.com/mhx/dwarfs
$ cd dwarfs

Once all dependencies have been installed, you can build DwarFS using:

$ mkdir build
$ cd build
$ cmake .. -GNinja -DWITH_TESTS=ON
$ ninja

You can then run tests with:

$ ctest -j

All binaries use jemalloc as a memory allocator by default, as it typically uses much less system memory compared to the glibc or tcmalloc allocators. To disable the use of jemalloc, pass -DUSE_JEMALLOC=0 on the cmake command line.

It is also possible to build/install the DwarFS libraries, tools, and FUSE driver independently. This is mostly interesting when packaging DwarFS. Note that the tools and FUSE driver require the libraries to be either built or already installed. To build just the libraries, use:

$ cmake .. -GNinja -DWITH_TESTS=ON -DWITH_LIBDWARFS=ON -DWITH_TOOLS=OFF -DWITH_FUSE_DRIVER=OFF

Once the libraries are tested and installed, you can build the tools (i.e. mkdwarfs, dwarfsck, dwarfsextract) using:

$ cmake .. -GNinja -DWITH_TESTS=ON -DWITH_LIBDWARFS=OFF -DWITH_TOOLS=ON -DWITH_FUSE_DRIVER=OFF

To build the FUSE driver, use:

$ cmake .. -GNinja -DWITH_TESTS=ON -DWITH_LIBDWARFS=OFF -DWITH_TOOLS=OFF -DWITH_FUSE_DRIVER=ON

Installing

Installing is as easy as:

$ sudo ninja install

Though you don’t have to install the tools to play with them.

Static Builds

Attempting to build statically linked binaries is highly discouraged and not officially supported. That being said, here’s how to set up an environment where you might be able to build static binaries.

This has been tested with ubuntu-22.04-live-server-amd64.iso. First, install all the packages listed as dependencies above. Also install:

$ apt install ccache ninja libacl1-dev

ccache and ninja are optional, but help with a speedy compile.

Depending on your distribution, you’ll need to build and install static versions of some libraries, e.g. libarchive and libmagic for Ubuntu:

$ wget https://github.com/libarchive/libarchive/releases/download/v3.6.2/libarchive-3.6.2.tar.xz
$ tar xf libarchive-3.6.2.tar.xz && cd libarchive-3.6.2
$ ./configure --prefix=/opt/static-libs --without-iconv --without-xml2 --without-expat
$ make && sudo make install

$ wget ftp://ftp.astron.com/pub/file/file-5.44.tar.gz
$ tar xf file-5.44.tar.gz && cd file-5.44
$ ./configure --prefix=/opt/static-libs --enable-static=yes --enable-shared=no
$ make && make install

That’s it! Now you can try building static binaries for DwarFS:

$ git clone --recurse-submodules https://github.com/mhx/dwarfs
$ cd dwarfs && mkdir build && cd build
$ cmake .. -GNinja -DWITH_TESTS=ON -DSTATIC_BUILD_DO_NOT_USE=ON \
           -DSTATIC_BUILD_EXTRA_PREFIX=/opt/static-libs
$ ninja
$ ninja test

Usage

Please check out the manual pages for mkdwarfs, dwarfs, dwarfsck and dwarfsextract. You can also access the manual pages using the --man option to each binary, e.g.:

$ mkdwarfs --man

The dwarfs manual page also shows an example for setting up DwarFS with overlayfs in order to create a writable file system mount on top of a read-only DwarFS image.

A description of the DwarFS file system format can be found in dwarfs-format(5).

A high-level overview of the internal operation of mkdwarfs is shown in this sequence diagram.

Using the Libraries

Using the DwarFS libraries should be pretty straightforward if you’re using CMake to build your project. For a quick start, have a look at the example code that uses the libraries to print information about a DwarFS image (like dwarfsck) or extract it (like dwarfsextract).

There are five individual libraries:

• dwarfs_common contains the common code required by all the other libraries. The interfaces are defined in dwarfs/.

• dwarfs_reader contains all code required to read data from a DwarFS image. The interfaces are defined in dwarfs/reader/.

• dwarfs_extractor contains the code required to extract a DwarFS image using libarchive. The interfaces are defined in dwarfs/utility/filesystem_extractor.h.

• dwarfs_writer contains the code required to create DwarFS images. The interfaces are defined in dwarfs/writer/.

• dwarfs_rewrite contains the code to re-write DwarFS images. The interfaces are defined in dwarfs/utility/rewrite_filesystem.h.

The headers in internal subfolders are only accessible at build time and won’t be installed. The same goes for the tool subfolder.

The reader and extractor APIs should be fairly stable. The writer APIs are likely going to change. Note, however, that there are no guarantees on API stability before this project reaches version 1.0.0.

Windows Support

Support for the Windows operating system is currently experimental. Having worked pretty much exclusively in a Unix world for the past two decades, my experience with Windows development is rather limited and I’d expect there to definitely be bugs and rough edges in the Windows code.

The Windows version of the DwarFS file system driver relies on the awesome WinFsp project and its winfsp-x64.dll must be discoverable by the dwarfs.exe driver.

The different tools should behave pretty much the same whether you’re using them on Linux or Windows. The file system images can be copied between Linux and Windows and images created on one OS should work fine on the other.

There are a few things worth pointing out, though:

• Windows traditionally treats file names as case-insensitive. However, the DwarFS driver is case-sensitive by default, even on Windows. There’s an option -o case_insensitive that can be passed to the driver to make it behave like a traditional Windows file system. Note that this is only supported if the DwarFS image contains no directories or files with the same name differing only in case within the same directory. Otherwise, the driver will emit a warning when mounting the image.

• DwarFS supports both hardlinks and symlinks on Windows, just as it does on Linux. However, creating hardlinks and symlinks seems to require admin privileges on Windows, so if, for example, you want to extract a DwarFS image that contains links of some sort, you might run into errors if you don’t have the right privileges.

• Due to a problem in WinFsp, symlinks cannot currently point outside of the mounted file system. Furthermore, due to another problem in WinFsp, symlinks with a drive letter will appear with a mangled target path.

• The DwarFS driver on Windows correctly reports hardlink counts via its API, but currently these counts are not correctly propagated to the Windows file system layer. This is presumably due to a problem in WinFsp.

• When mounting a DwarFS image on Windows, the mount point must not exist. This is different from Linux, where the mount point must actually exist. Also, it’s possible to mount a DwarFS image as a drive letter, e.g.

  dwarfs.exe image.dwarfs Z:

• Filter rules for mkdwarfs always require Unix path separators, regardless of whether it’s running on Windows or Linux.

Building on Windows

Building on Windows is not too complicated thanks to vcpkg. You’ll need to install:

• Visual Studio and the MSVC C/C++ compiler

• Git

• CMake

• Ninja

• WinFsp

WinFsp is expected to be installed in C:\Program Files (x86)\WinFsp; if it’s not, you’ll need to set WINFSP_PATH when running CMake via cmake/win.bat.

Clone vcpkg and dwarfs:

> cd %HOMEPATH%
> mkdir git
> cd git
> git clone https://github.com/Microsoft/vcpkg.git
> git clone https://github.com/mhx/dwarfs

Then, bootstrap vcpkg:

> .\vcpkg\bootstrap-vcpkg.bat

And build DwarFS:

> cd dwarfs
> mkdir build
> cd build
> ..\cmake\win.bat
> ninja

Once that’s done, you should be able to run the tests. Set CTEST_PARALLEL_LEVEL according to the number of CPU cores in your machine.

> set CTEST_PARALLEL_LEVEL=10
> ninja test

macOS Support

The DwarFS libraries and tools (mkdwarfs, dwarfsck, dwarfsextract) are now available from Homebrew:

$ brew install dwarfs
$ brew test dwarfs

The macOS version of the DwarFS file system driver relies on the awesome macFUSE project and is available via gromgit’s homebrew-fuse tap:

$ brew tap gromgit/homebrew-fuse
$ brew install dwarfs-fuse-mac

Occasionally, when the main Homebrew repository switches to newer library versions, the bottles in the homebrew-fuse tap can break. In that case, you will see the FUSE driver fail to run with errors about missing libraries from dyld. If that happens, you can file an issue for the homebrew-fuse tap. You can also temporarily work around the problem by building the FUSE driver from source:

$ brew reinstall --build-from-source dwarfs-fuse-mac

Use Cases

Astrophotography

Astrophotography can generate huge amounts of raw image data. During a single night, it’s not unlikely to end up with a few dozen gigabytes of data. With most dedicated astrophotography cameras, this data ends up in the form of FITS images. These are usually uncompressed, don’t compress very well with standard compression algorithms, and while there are certain compressed FITS formats, these aren’t widely supported.

One of the compression formats (simply called “Rice”) compresses reasonably well and is really fast. However, its implementation for compressed FITS has a few drawbacks. The most severe drawbacks are that compression isn’t quite as good as it could be for color sensors and sensors with a less than 16 bits of resolution.

DwarFS supports the ricepp (Rice++) compression, which builds on the basic idea of Rice compression, but makes a few enhancements: it compresses color and low bit depth images significantly better and always searches for the optimum solution during compression instead of relying on a heuristic.

Let’s look at an example using 129 images (darks, flats and lights) taken with an ASI1600MM camera. Each image is 32 MiB, so a total of 4 GiB of data. Compressing these with the standard fpack tool takes about 16.6 seconds and yields a total output size of 2.2 GiB:

$ time fpack */*.fit */*/*.fit

user    14.992
system  1.592
total   16.616

$ find . -name '*.fz' -print0 | xargs -0 cat | wc -c
2369943360

However, this leaves you with *.fz files that not every application can actually read.

Using DwarFS, here’s what we get:

$ mkdwarfs -i ASI1600 -o asi1600-20.dwarfs -S 20 --categorize
I 08:47:47.459077 scanning "ASI1600"
I 08:47:47.491492 assigning directory and link inodes...
I 08:47:47.491560 waiting for background scanners...
I 08:47:47.675241 scanning CPU time: 1.051s
I 08:47:47.675271 finalizing file inodes...
I 08:47:47.675330 saved 0 B / 3.941 GiB in 0/258 duplicate files
I 08:47:47.675360 assigning device inodes...
I 08:47:47.675371 assigning pipe/socket inodes...
I 08:47:47.675381 building metadata...
I 08:47:47.675393 building blocks...
I 08:47:47.675398 saving names and symlinks...
I 08:47:47.675514 updating name and link indices...
I 08:47:47.675796 waiting for segmenting/blockifying to finish...
I 08:47:50.274285 total ordering CPU time: 616.3us
I 08:47:50.274329 total segmenting CPU time: 1.132s
I 08:47:50.279476 saving chunks...
I 08:47:50.279622 saving directories...
I 08:47:50.279674 saving shared files table...
I 08:47:50.280745 saving names table... [1.047ms]
I 08:47:50.280768 saving symlinks table... [743ns]
I 08:47:50.282031 waiting for compression to finish...
I 08:47:50.823924 compressed 3.941 GiB to 1.201 GiB (ratio=0.304825)
I 08:47:50.824280 compression CPU time: 17.92s
I 08:47:50.824316 filesystem created without errors [3.366s]
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
waiting for block compression to finish
5 dirs, 0/0 soft/hard links, 258/258 files, 0 other
original size: 3.941 GiB, hashed: 315.4 KiB (18 files, 0 B/s)
scanned: 3.941 GiB (258 files, 117.1 GiB/s), categorizing: 0 B/s
saved by deduplication: 0 B (0 files), saved by segmenting: 0 B
filesystem: 3.941 GiB in 4037 blocks (4550 chunks, 516/516 fragments, 258 inodes)
compressed filesystem: 4037 blocks/1.201 GiB written

In less than 3.4 seconds, it compresses the data down to 1.2 GiB, almost half the size of the fpack output.

In addition to saving a lot of disk space, this can also be useful when your data is stored on a NAS. Here’s a comparison of the same set of data accessed over a 1 Gb/s network connection, first using the uncompressed raw data:

find /mnt/ASI1600 -name '*.fit' -print0 | xargs -0 -P4 -n1 cat | dd of=/dev/null status=progress
4229012160 bytes (4.2 GB, 3.9 GiB) copied, 36.0455 s, 117 MB/s

And next using a DwarFS image on the same share:

$ dwarfs /mnt/asi1600-20.dwarfs mnt

$ find mnt -name '*.fit' -print0 | xargs -0 -P4 -n1 cat | dd of=/dev/null status=progress
4229012160 bytes (4.2 GB, 3.9 GiB) copied, 14.3681 s, 294 MB/s

That’s roughly 2.5 times faster. You can very likely see similar results with slow external hard drives.

Dealing with Bit Rot

Currently, DwarFS has no built-in ability to add recovery information to a file system image. However, for archival purposes, it’s a good idea to have such recovery information in order to be able to repair a damaged image.

This is fortunately relatively straightforward using something like par2cmdline:

$ par2create -n1 asi1600-20.dwarfs

This will create two additional files that you can place alongside the image (or on a different storage), as you’ll only need them if DwarFS has detected an issue with the file system image. If there’s an issue, you can run

$ par2repair asi1600-20.dwarfs

which will very likely be able to recover the image if less than 5% (that’s the default used by par2create) of the image are damaged.

Extended Attributes

Preserving Extended Attributes in DwarFS Images

Extended attributes are not currently supported. Any extended attributes stored in the source file system will not currently be preserved when building a DwarFS image using mkdwarfs.

Extended Attributes exposed by the FUSE Driver

That being said, the root inode of a mounted DwarFS image currently exposes one or two extended attributes on Linux:

$ attr -l mnt
Attribute "dwarfs.driver.pid" has a 4 byte value for mnt
Attribute "dwarfs.driver.perfmon" has a 4849 byte value for mnt

The dwarfs.driver.pid attribute simply contains the PID of the DwarFS FUSE driver. The dwarfs.driver.perfmon attribute contains the current results of the performance monitor.

Furthermore, each regular file exposes an attribute dwarfs.inodeinfo with information about the underlying inode:

$ attr -l "05 Disappear.caf"
Attribute "dwarfs.inodeinfo" has a 448 byte value for 05 Disappear.caf

The attribute contains a JSON object with information about the underlying inode:

$ attr -qg dwarfs.inodeinfo "05 Disappear.caf"
{
  "chunks": [
    {
      "block": 2,
      "category": "pcmaudio/metadata",
      "offset": 270976,
      "size": 4096
    },
    {
      "block": 414,
      "category": "pcmaudio/waveform",
      "offset": 37594368,
      "size": 29514492
    },
    {
      "block": 419,
      "category": "pcmaudio/waveform",
      "offset": 0,
      "size": 29385468
    }
  ],
  "gid": 100,
  "mode": 33188,
  "modestring": "----rw-r--r--",
  "uid": 1000
}

This is useful, for example, to check how a particular file is spread across multiple blocks or which categories have been assigned to the file.

Comparison

The SquashFS, xz, lrzip, zpaq and wimlib tests were all done on an 8 core Intel® Xeon® E-2286M CPU @ 2.40GHz with 64 GiB of RAM.

The Cromfs tests were done with an older version of DwarFS on a 6 core Intel® Xeon® CPU D-1528 @ 1.90GHz with 64 GiB of RAM.

The EROFS tests were done using DwarFS v0.9.8 and EROFS v1.7.1 on an Intel® Core™ i9-13900K with 64 GiB of RAM.

The systems were mostly idle during all of the tests.

With SquashFS

The source directory contained 1139 different Perl installations from 284 distinct releases, a total of 47.65 GiB of data in 1,927,501 files and 330,733 directories. The source directory was freshly unpacked from a tar archive to an XFS partition on a 970 EVO Plus 2TB NVMe drive, so most of its contents were likely cached.

I’m using the same compression type and compression level for SquashFS that is the default setting for DwarFS:

$ time mksquashfs install perl-install.squashfs -comp zstd -Xcompression-level 22
Parallel mksquashfs: Using 16 processors
Creating 4.0 filesystem on perl-install-zstd.squashfs, block size 131072.
[=========================================================/] 2107401/2107401 100%

Exportable Squashfs 4.0 filesystem, zstd compressed, data block size 131072
        compressed data, compressed metadata, compressed fragments,
        compressed xattrs, compressed ids
        duplicates are removed
Filesystem size 4637597.63 Kbytes (4528.90 Mbytes)
        9.29% of uncompressed filesystem size (49922299.04 Kbytes)
Inode table size 19100802 bytes (18653.13 Kbytes)
        26.06% of uncompressed inode table size (73307702 bytes)
Directory table size 19128340 bytes (18680.02 Kbytes)
        46.28% of uncompressed directory table size (41335540 bytes)
Number of duplicate files found 1780387
Number of inodes 2255794
Number of files 1925061
Number of fragments 28713
Number of symbolic links  0
Number of device nodes 0
Number of fifo nodes 0
Number of socket nodes 0
Number of directories 330733
Number of ids (unique uids + gids) 2
Number of uids 1
        mhx (1000)
Number of gids 1
        users (100)

real    32m54.713s
user    501m46.382s
sys     0m58.528s

For DwarFS, I’m sticking to the defaults:

$ time mkdwarfs -i install -o perl-install.dwarfs
I 11:33:33.310931 scanning install
I 11:33:39.026712 waiting for background scanners...
I 11:33:50.681305 assigning directory and link inodes...
I 11:33:50.888441 finding duplicate files...
I 11:34:01.120800 saved 28.2 GiB / 47.65 GiB in 1782826/1927501 duplicate files
I 11:34:01.122608 waiting for inode scanners...
I 11:34:12.839065 assigning device inodes...
I 11:34:12.875520 assigning pipe/socket inodes...
I 11:34:12.910431 building metadata...
I 11:34:12.910524 building blocks...
I 11:34:12.910594 saving names and links...
I 11:34:12.910691 bloom filter size: 32 KiB
I 11:34:12.910760 ordering 144675 inodes using nilsimsa similarity...
I 11:34:12.915555 nilsimsa: depth=20000 (1000), limit=255
I 11:34:13.052525 updating name and link indices...
I 11:34:13.276233 pre-sorted index (660176 name, 366179 path lookups) [360.6ms]
I 11:35:44.039375 144675 inodes ordered [91.13s]
I 11:35:44.041427 waiting for segmenting/blockifying to finish...
I 11:37:38.823902 bloom filter reject rate: 96.017% (TPR=0.244%, lookups=4740563665)
I 11:37:38.823963 segmentation matches: good=454708, bad=6819, total=464247
I 11:37:38.824005 segmentation collisions: L1=0.008%, L2=0.000% [2233254 hashes]
I 11:37:38.824038 saving chunks...
I 11:37:38.860939 saving directories...
I 11:37:41.318747 waiting for compression to finish...
I 11:38:56.046809 compressed 47.65 GiB to 430.9 MiB (ratio=0.00883101)
I 11:38:56.304922 filesystem created without errors [323s]
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
waiting for block compression to finish
330733 dirs, 0/2440 soft/hard links, 1927501/1927501 files, 0 other
original size: 47.65 GiB, dedupe: 28.2 GiB (1782826 files), segment: 15.19 GiB
filesystem: 4.261 GiB in 273 blocks (319178 chunks, 144675/144675 inodes)
compressed filesystem: 273 blocks/430.9 MiB written [depth: 20000]
█████████████████████████████████████████████████████████████████████████████▏100% |

real    5m23.030s
user    78m7.554s
sys     1m47.968s

So in this comparison, mkdwarfs is more than 6 times faster than mksquashfs, both in terms of CPU time and wall clock time.

$ ll perl-install.*fs
-rw-r--r-- 1 mhx users  447230618 Mar  3 20:28 perl-install.dwarfs
-rw-r--r-- 1 mhx users 4748902400 Mar  3 20:10 perl-install.squashfs

In terms of compression ratio, the DwarFS file system is more than 10 times smaller than the SquashFS file system. With DwarFS, the content has been compressed down to less than 0.9% (!) of its original size. This compression ratio only considers the data stored in the individual files, not the actual disk space used. On the original XFS file system, according to du, the source folder uses 52 GiB, so the DwarFS image actually only uses 0.8% of the original space.

Here’s another comparison using lzma compression instead of zstd:

$ time mksquashfs install perl-install-lzma.squashfs -comp lzma

real    13m42.825s
user    205m40.851s
sys     3m29.088s

$ time mkdwarfs -i install -o perl-install-lzma.dwarfs -l9

real    3m43.937s
user    49m45.295s
sys     1m44.550s

$ ll perl-install-lzma.*fs
-rw-r--r-- 1 mhx users  315482627 Mar  3 21:23 perl-install-lzma.dwarfs
-rw-r--r-- 1 mhx users 3838406656 Mar  3 20:50 perl-install-lzma.squashfs

It’s immediately obvious that the runs are significantly faster and the resulting images are significantly smaller. Still, mkdwarfs is about 4 times faster and produces and image that’s 12 times smaller than the SquashFS image. The DwarFS image is only 0.6% of the original file size.

So, why not use lzma instead of zstd by default? The reason is that lzma is about an order of magnitude slower to decompress than zstd. If you’re only accessing data on your compressed file system occasionally, this might not be a big deal, but if you use it extensively, zstd will result in better performance.

The comparisons above are not completely fair. mksquashfs by default uses a block size of 128KiB, whereas mkdwarfs uses 16MiB blocks by default, or even 64MiB blocks with -l9. When using identical block sizes for both file systems, the difference, quite expectedly, becomes a lot less dramatic:

$ time mksquashfs install perl-install-lzma-1M.squashfs -comp lzma -b 1M

real    15m43.319s
user    139m24.533s
sys     0m45.132s

$ time mkdwarfs -i install -o perl-install-lzma-1M.dwarfs -l9 -S20 -B3

real    4m25.973s
user    52m15.100s
sys     7m41.889s

$ ll perl-install*.*fs
-rw-r--r-- 1 mhx users  935953866 Mar 13 12:12 perl-install-lzma-1M.dwarfs
-rw-r--r-- 1 mhx users 3407474688 Mar  3 21:54 perl-install-lzma-1M.squashfs

Even this is still not entirely fair, as it uses a feature (-B3) that allows DwarFS to reference file chunks from up to two previous file system blocks.

But the point is that this is really where SquashFS tops out, as it doesn’t support larger block sizes or back-referencing. And as you’ll see below, the larger blocks that DwarFS is using by default don’t necessarily negatively impact performance.

DwarFS also features an option to recompress an existing file system with a different compression algorithm. This can be useful as it allows relatively fast experimentation with different algorithms and options without requiring a full rebuild of the file system. For example, recompressing the above file system with the best possible compression (-l 9):

$ time mkdwarfs --recompress -i perl-install.dwarfs -o perl-lzma-re.dwarfs -l9
I 20:28:03.246534 filesystem rewritten without errors [148.3s]
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
filesystem: 4.261 GiB in 273 blocks (0 chunks, 0 inodes)
compressed filesystem: 273/273 blocks/372.7 MiB written
████████████████████████████████████████████████████████████████████▏100% \

real    2m28.279s
user    37m8.825s
sys     0m43.256s

$ ll perl-*.dwarfs
-rw-r--r-- 1 mhx users 447230618 Mar  3 20:28 perl-install.dwarfs
-rw-r--r-- 1 mhx users 390845518 Mar  4 20:28 perl-lzma-re.dwarfs
-rw-r--r-- 1 mhx users 315482627 Mar  3 21:23 perl-install-lzma.dwarfs

Note that while the recompressed file system is smaller than the original image, it is still a lot bigger than the file system we previously build with -l9. The reason is that the recompressed image still uses the same block size, and the block size cannot be changed by recompressing.

In terms of how fast the file system is when using it, a quick test I’ve done is to freshly mount the file system created above and run each of the 1139 perl executables to print their version.

$ hyperfine -c "umount mnt" -p "umount mnt; dwarfs perl-install.dwarfs mnt -o cachesize=1g -o workers=4; sleep 1" -P procs 5 20 -D 5 "ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P{procs} sh -c '\$0 -v >/dev/null'"
Benchmark #1: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P5 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      1.810 s ±  0.013 s    [User: 1.847 s, System: 0.623 s]
  Range (min … max):    1.788 s …  1.825 s    10 runs

Benchmark #2: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      1.333 s ±  0.009 s    [User: 1.993 s, System: 0.656 s]
  Range (min … max):    1.321 s …  1.354 s    10 runs

Benchmark #3: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P15 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      1.181 s ±  0.018 s    [User: 2.086 s, System: 0.712 s]
  Range (min … max):    1.165 s …  1.214 s    10 runs

Benchmark #4: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      1.149 s ±  0.015 s    [User: 2.128 s, System: 0.781 s]
  Range (min … max):    1.136 s …  1.186 s    10 runs

These timings are for initial runs on a freshly mounted file system, running 5, 10, 15 and 20 processes in parallel. 1.1 seconds means that it takes only about 1 millisecond per Perl binary.

Following are timings for subsequent runs, both on DwarFS (at mnt) and the original XFS (at install). DwarFS is around 15% slower here:

$ hyperfine -P procs 10 20 -D 10 -w1 "ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P{procs} sh -c '\$0 -v >/dev/null'" "ls -1 install/*/*/bin/perl5* | xargs -d $'\n' -n1 -P{procs} sh -c '\$0 -v >/dev/null'"
Benchmark #1: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):     347.0 ms ±   7.2 ms    [User: 1.755 s, System: 0.452 s]
  Range (min … max):   341.3 ms … 365.2 ms    10 runs

Benchmark #2: ls -1 install/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):     302.5 ms ±   3.3 ms    [User: 1.656 s, System: 0.377 s]
  Range (min … max):   297.1 ms … 308.7 ms    10 runs

Benchmark #3: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):     342.2 ms ±   4.1 ms    [User: 1.766 s, System: 0.451 s]
  Range (min … max):   336.0 ms … 349.7 ms    10 runs

Benchmark #4: ls -1 install/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):     302.0 ms ±   3.0 ms    [User: 1.659 s, System: 0.374 s]
  Range (min … max):   297.0 ms … 305.4 ms    10 runs

Summary
  'ls -1 install/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null'' ran
    1.00 ± 0.01 times faster than 'ls -1 install/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null''
    1.13 ± 0.02 times faster than 'ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null''
    1.15 ± 0.03 times faster than 'ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null''

Using the lzma-compressed file system, the metrics for initial runs look considerably worse (about an order of magnitude):

$ hyperfine -c "umount mnt" -p "umount mnt; dwarfs perl-install-lzma.dwarfs mnt -o cachesize=1g -o workers=4; sleep 1" -P procs 5 20 -D 5 "ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P{procs} sh -c '\$0 -v >/dev/null'"
Benchmark #1: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P5 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):     10.660 s ±  0.057 s    [User: 1.952 s, System: 0.729 s]
  Range (min … max):   10.615 s … 10.811 s    10 runs

Benchmark #2: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P10 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      9.092 s ±  0.021 s    [User: 1.979 s, System: 0.680 s]
  Range (min … max):    9.059 s …  9.126 s    10 runs

Benchmark #3: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P15 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      9.012 s ±  0.188 s    [User: 2.077 s, System: 0.702 s]
  Range (min … max):    8.839 s …  9.277 s    10 runs

Benchmark #4: ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '$0 -v >/dev/null'
  Time (mean ± σ):      9.004 s ±  0.298 s    [User: 2.134 s, System: 0.736 s]
  Range (min … max):    8.611 s …  9.555 s    10 runs

So you might want to consider using zstd instead of lzma if you’d like to optimize for file system performance. It’s also the default compression used by mkdwarfs.

Now here’s a comparison with the SquashFS file system:

$ hyperfine -c 'sudo umount mnt' -p 'umount mnt; dwarfs perl-install.dwarfs mnt -o cachesize=1g -o workers=4; sleep 1' -n dwarfs-zstd "ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '\$0 -v >/dev/null'" -p 'sudo umount mnt; sudo mount -t squashfs perl-install.squashfs mnt; sleep 1' -n squashfs-zstd "ls -1 mnt/*/*/bin/perl5* | xargs -d $'\n' -n1 -P20 sh -c '\$0 -v >/dev/null'"
Benchmark #1: dwarfs-zstd
  Time (mean ± σ):      1.151 s ±  0.015 s    [User: 2.147 s, System: 0.769 s]
  Range (min … max):    1.118 s …  1.174 s    10 runs

Benchmark #2: squashfs-zstd
  Time (mean ± σ):      6.733 s ±  0.007 s    [User: 3.188 s, System: 17.015 s]
  Range (min … max):    6.721 s …  6.743 s    10 runs

Summary
  'dwarfs-zstd' ran
    5.85 ± 0.08 times faster than 'squashfs-zstd'

So, DwarFS is almost six times faster than SquashFS. But what’s more, SquashFS also uses significantly more CPU power. However, the numbers shown above for DwarFS obviously don’t include the time spent in the dwarfs process, so I repeated the test outside of hyperfine:

$ time dwarfs perl-install.dwarfs mnt -o cachesize=1g -o workers=4 -f

real    0m4.569s
user    0m2.154s
sys     0m1.846s

So, in total, DwarFS was using 5.7 seconds of CPU time, whereas SquashFS was using 20.2 seconds, almost four times as much. Ignore the ‘real’ time, this is only how long it took me to unmount the file system again after mounting it.

Another real-life test was to build and test a Perl module with 624 different Perl versions in the compressed file system. The module I’ve used, Tie::Hash::Indexed, has an XS component that requires a C compiler to build. So this really accesses a lot of different stuff in the file system:

• The perl executables and its shared libraries

• The Perl modules used for writing the Makefile

• Perl’s C header files used for building the module

• More Perl modules used for running the tests

I wrote a little script to be able to run multiple builds in parallel:

#!/bin/bash
set -eu
perl=$1
dir=$(echo "$perl" | cut -d/ --output-delimiter=- -f5,6)
rsync -a Tie-Hash-Indexed/ $dir/
cd $dir
$1 Makefile.PL >/dev/null 2>&1
make test >/dev/null 2>&1
cd ..
rm -rf $dir
echo $perl

The following command will run up to 16 builds in parallel on the 8 core Xeon CPU, including debug, optimized and threaded versions of all Perl releases between 5.10.0 and 5.33.3, a total of 624 perl installations:

$ time ls -1 /tmp/perl/install/*/perl-5.??.?/bin/perl5* | sort -t / -k 8 | xargs -d $'\n' -P 16 -n 1 ./build.sh

Tests were done with a cleanly mounted file system to make sure the caches were empty. ccache was primed to make sure all compiler runs could be satisfied from the cache. With SquashFS, the timing was:

real    0m52.385s
user    8m10.333s
sys     4m10.056s

And with DwarFS:

real    0m50.469s
user    9m22.597s
sys     1m18.469s

So, frankly, not much of a difference, with DwarFS being just a bit faster. The dwarfs process itself used:

real    0m56.686s
user    0m18.857s
sys     0m21.058s

So again, DwarFS used less raw CPU power overall, but in terms of wallclock time, the difference is really marginal.

With SquashFS & xz

This test uses slightly less pathological input data: the root file system of a recent Raspberry Pi OS release. This file system also contains device inodes, so in order to preserve those, we pass --with-devices to mkdwarfs:

$ time sudo mkdwarfs -i raspbian -o raspbian.dwarfs --with-devices
I 21:30:29.812562 scanning raspbian
I 21:30:29.908984 waiting for background scanners...
I 21:30:30.217446 assigning directory and link inodes...
I 21:30:30.221941 finding duplicate files...
I 21:30:30.288099 saved 31.05 MiB / 1007 MiB in 1617/34582 duplicate files
I 21:30:30.288143 waiting for inode scanners...
I 21:30:31.393710 assigning device inodes...
I 21:30:31.394481 assigning pipe/socket inodes...
I 21:30:31.395196 building metadata...
I 21:30:31.395230 building blocks...
I 21:30:31.395291 saving names and links...
I 21:30:31.395374 ordering 32965 inodes using nilsimsa similarity...
I 21:30:31.396254 nilsimsa: depth=20000 (1000), limit=255
I 21:30:31.407967 pre-sorted index (46431 name, 2206 path lookups) [11.66ms]
I 21:30:31.410089 updating name and link indices...
I 21:30:38.178505 32965 inodes ordered [6.783s]
I 21:30:38.179417 waiting for segmenting/blockifying to finish...
I 21:31:06.248304 saving chunks...
I 21:31:06.251998 saving directories...
I 21:31:06.402559 waiting for compression to finish...
I 21:31:16.425563 compressed 1007 MiB to 287 MiB (ratio=0.285036)
I 21:31:16.464772 filesystem created without errors [46.65s]
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
waiting for block compression to finish
4435 dirs, 5908/0 soft/hard links, 34582/34582 files, 7 other
original size: 1007 MiB, dedupe: 31.05 MiB (1617 files), segment: 47.23 MiB
filesystem: 928.4 MiB in 59 blocks (38890 chunks, 32965/32965 inodes)
compressed filesystem: 59 blocks/287 MiB written [depth: 20000]
████████████████████████████████████████████████████████████████████▏100% |

real    0m46.711s
user    10m39.038s
sys     0m8.123s

Again, SquashFS uses the same compression options:

$ time sudo mksquashfs raspbian raspbian.squashfs -comp zstd -Xcompression-level 22
Parallel mksquashfs: Using 16 processors
Creating 4.0 filesystem on raspbian.squashfs, block size 131072.
[===============================================================\] 39232/39232 100%

Exportable Squashfs 4.0 filesystem, zstd compressed, data block size 131072
        compressed data, compressed metadata, compressed fragments,
        compressed xattrs, compressed ids
        duplicates are removed
Filesystem size 371934.50 Kbytes (363.22 Mbytes)
        35.98% of uncompressed filesystem size (1033650.60 Kbytes)
Inode table size 399913 bytes (390.54 Kbytes)
        26.53% of uncompressed inode table size (1507581 bytes)
Directory table size 408749 bytes (399.17 Kbytes)
        42.31% of uncompressed directory table size (966174 bytes)
Number of duplicate files found 1618
Number of inodes 44932
Number of files 34582
Number of fragments 3290
Number of symbolic links  5908
Number of device nodes 7
Number of fifo nodes 0
Number of socket nodes 0
Number of directories 4435
Number of ids (unique uids + gids) 18
Number of uids 5
        root (0)
        mhx (1000)
        unknown (103)
        shutdown (6)
        unknown (106)
Number of gids 15
        root (0)
        unknown (109)
        unknown (42)
        unknown (1000)
        users (100)
        unknown (43)
        tty (5)
        unknown (108)
        unknown (111)
        unknown (110)
        unknown (50)
        mail (12)
        nobody (65534)
        adm (4)
        mem (8)

real    0m50.124s
user    9m41.708s
sys     0m1.727s

The difference in speed is almost negligible. SquashFS is just a bit slower here. In terms of compression, the difference also isn’t huge:

$ ls -lh raspbian.* *.xz
-rw-r--r-- 1 mhx  users 297M Mar  4 21:32 2020-08-20-raspios-buster-armhf-lite.img.xz
-rw-r--r-- 1 root root  287M Mar  4 21:31 raspbian.dwarfs
-rw-r--r-- 1 root root  364M Mar  4 21:33 raspbian.squashfs

Interestingly, xz actually can’t compress the whole original image better than DwarFS.

We can even again try to increase the DwarFS compression level:

$ time sudo mkdwarfs -i raspbian -o raspbian-9.dwarfs --with-devices -l9

real    0m54.161s
user    8m40.109s
sys     0m7.101s

Now that actually gets the DwarFS image size well below that of the xz archive:

$ ls -lh raspbian-9.dwarfs *.xz
-rw-r--r-- 1 root root  244M Mar  4 21:36 raspbian-9.dwarfs
-rw-r--r-- 1 mhx  users 297M Mar  4 21:32 2020-08-20-raspios-buster-armhf-lite.img.xz

Even if you actually build a tarball and compress that (instead of compressing the EXT4 file system itself), xz isn’t quite able to match the DwarFS image size:

$ time sudo tar cf - raspbian | xz -9 -vT 0 >raspbian.tar.xz
  100 %     246.9 MiB / 1,037.2 MiB = 0.238    13 MiB/s       1:18

real    1m18.226s
user    6m35.381s
sys     0m2.205s

$ ls -lh raspbian.tar.xz
-rw-r--r-- 1 mhx users 247M Mar  4 21:40 raspbian.tar.xz

DwarFS also comes with the dwarfsextract tool that allows extraction of a file system image without the FUSE driver. So here’s a comparison of the extraction speed:

$ time sudo tar xf raspbian.tar.xz -C out1

real    0m12.846s
user    0m12.313s
sys     0m1.616s

$ time sudo dwarfsextract -i raspbian-9.dwarfs -o out2

real    0m3.825s
user    0m13.234s
sys     0m1.382s

So, dwarfsextract is almost 4 times faster thanks to using multiple worker threads for decompression. It’s writing about 300 MiB/s in this example.

Another nice feature of dwarfsextract is that it allows you to directly output data in an archive format, so you could create a tarball from your image without extracting the files to disk:

$ dwarfsextract -i raspbian-9.dwarfs -f ustar | xz -9 -T0 >raspbian2.tar.xz

This has the interesting side-effect that the resulting tarball will likely be smaller than the one built straight from the directory:

$ ls -lh raspbian*.tar.xz
-rw-r--r-- 1 mhx users 247M Mar  4 21:40 raspbian.tar.xz
-rw-r--r-- 1 mhx users 240M Mar  4 23:52 raspbian2.tar.xz

That’s because dwarfsextract writes files in inode-order, and by default inodes are ordered by similarity for the best possible compression.

With lrzip

lrzip is a compression utility targeted e

README 内容过长，已截取前 64 KB。完整内容请前往 GitHub 仓库查看。