3 This package implements an "erasure code", or "forward error correction code".
5 You may use this package under the GNU General Public License, version 2 or, at
6 your option, any later version. You may use this package under the Transitive
7 Grace Period Public Licence, version 1.0. (You may choose to use this package
8 under the terms of either licence, at your option.) See the file COPYING.GPL
9 for the terms of the GNU General Public License, version 2. See the file
10 COPYING.TGPPL.html for the terms of the Transitive Grace Period Public Licence,
11 version 1.0. In addition, Allmydata, Inc. offers other licensing terms. If you
12 would like to inquire about a commercial relationship with Allmydata, Inc.,
13 please contact partnerships@allmydata.com and visit http://allmydata.com .
15 The most widely known example of an erasure code is the RAID-5 algorithm which
16 makes it so that in the event of the loss of any one hard drive, the stored data
17 can be completely recovered. The algorithm in the zfec package has a similar
18 effect, but instead of recovering from the loss of only a single element, it can
19 be parameterized to choose in advance the number of elements whose loss it can
22 This package is largely based on the old "fec" library by Luigi Rizzo et al.,
23 which is a mature and optimized implementation of erasure coding. The zfec
24 package makes several changes from the original "fec" package, including
25 addition of the Python API, refactoring of the C API to support zero-copy
26 operation, a few clean-ups and optimizations of the core code itself, and the
27 addition of a command-line tool named "zfec".
32 This package is managed with the "setuptools" package management tool. To build
33 and install the package directly into your system, just run "python ./setup.py
34 install". If you prefer to keep the package limited to a specific directory so
35 that you can manage it yourself (perhaps by using the "GNU stow") tool, then
36 give it these arguments: "python ./setup.py install
37 --single-version-externally-managed
38 --record=${specificdirectory}/zfec-install.log --prefix=${specificdirectory}"
40 To run the self-tests, execute "python ./setup.py test" (or if you have Twisted
41 Python installed, you can run "trial zfec" for nicer output and test options.)
42 This will run the tests of the C API, the Python API, and the command-line
45 To run the tests of the Haskell API:
46 % runhaskell haskell/test/FECTest.hs
48 Note that in order to run the Haskell API tests you must have installed the
49 library first due to the fact that the interpreter cannot process FEC.hs as it
50 takes a reference to an FFI function.
55 The source is currently available via darcs on the web with the command:
57 darcs get http://allmydata.org/source/zfec/trunk
59 More information on darcs is available at http://darcs.net
61 Please join the zfec mailing list and submit patches:
63 <http://allmydata.org/cgi-bin/mailman/listinfo/zfec-dev>
68 This package performs two operations, encoding and decoding. Encoding takes
69 some input data and expands its size by producing extra "check blocks", also
70 called "secondary blocks". Decoding takes some data -- any combination of
71 blocks of the original data (called "primary blocks") and "secondary blocks",
72 and produces the original data.
74 The encoding is parameterized by two integers, k and m. m is the total number
75 of blocks produced, and k is how many of those blocks are necessary to
76 reconstruct the original data. m is required to be at least 1 and at most 256,
77 and k is required to be at least 1 and at most m.
79 (Note that when k == m then there is no point in doing erasure coding -- it
80 degenerates to the equivalent of the Unix "split" utility which simply splits
81 the input into successive segments. Similarly, when k == 1 it degenerates to
82 the equivalent of the unix "cp" utility -- each block is a complete copy of the
85 Note that each "primary block" is a segment of the original data, so its size is
86 1/k'th of the size of original data, and each "secondary block" is of the same
87 size, so the total space used by all the blocks is m/k times the size of the
88 original data (plus some padding to fill out the last primary block to be the
89 same size as all the others). In addition to the data contained in the blocks
90 themselves there are also a few pieces of metadata which are necessary for later
91 reconstruction. Those pieces are: 1. the value of K, 2. the value of M, 3.
92 the sharenum of each block, 4. the number of bytes of padding that were used.
93 The "zfec" command-line tool compresses these pieces of data and prepends them
94 to the beginning of each share, so each the sharefile produced by the "zfec"
95 command-line tool is between one and four bytes larger than the share data
98 The decoding step requires as input k of the blocks which were produced by the
99 encoding step. The decoding step produces as output the data that was earlier
100 input to the encoding step.
105 The bin/ directory contains two Unix-style, command-line tools "zfec" and
106 "zunfec". Execute "zfec --help" or "zunfec --help" for usage instructions.
108 Note: a Unix-style tool like "zfec" does only one thing -- in this case erasure
109 coding -- and leaves other tasks to other tools. Other Unix-style tools that go
110 well with zfec include "GNU tar" or "7z" a.k.a. "p7zip" for archiving multiple
111 files and directories into one file, "7z" or "rzip" for compression, and "GNU Privacy
112 Guard" for encryption or "sha256sum" for integrity. It is important to do
113 things in order: first archive, then compress, then either encrypt or sha256sum,
114 then erasure code. Note that if GNU Privacy Guard is used for privacy, then it
115 will also ensure integrity, so the use of sha256sum is unnecessary in that case.
116 Note that if 7z is used for archiving then it also does compression, so you
117 don't need a separate compressor in that case.
120 * Performance Measurements
122 On my Athlon 64 2.4 GHz workstation (running Linux), the "zfec" command-line
123 tool encoded a 160 MB file with m=100, k=94 (about 6% redundancy) in 3.9
124 seconds, where the "par2" tool encoded the file with about 6% redundancy in 27
125 seconds. zfec encoded the same file with m=12, k=6 (100% redundancy) in 4.1
126 seconds, where par2 encoded it with about 100% redundancy in 7 minutes and 56
129 The underlying C library in benchmark mode encoded from a file at about 4.9
130 million bytes per second and decoded at about 5.8 million bytes per second.
132 On Peter's fancy Intel Mac laptop (2.16 GHz Core Duo), it encoded from a file at
133 about 6.2 million bytes per second.
135 On my even fancier Intel Mac laptop (2.33 GHz Core Duo), it encoded from a file
136 at about 6.8 million bytes per second.
138 On my old PowerPC G4 867 MHz Mac laptop, it encoded from a file at about 1.3
139 million bytes per second.
144 Each block is associated with "blocknum". The blocknum of each primary block is
145 its index (starting from zero), so the 0'th block is the first primary block,
146 which is the first few bytes of the file, the 1'st block is the next primary
147 block, which is the next few bytes of the file, and so on. The last primary
148 block has blocknum k-1. The blocknum of each secondary block is an arbitrary
149 integer between k and 255 inclusive. (When using the Python API, if you don't
150 specify which secondary blocks you want when invoking encode(), then it will by
151 default provide the blocks with ids from k to m-1 inclusive.)
155 fec_encode() takes as input an array of k pointers, where each pointer points to
156 a memory buffer containing the input data (i.e., the i'th buffer contains the
157 i'th primary block). There is also a second parameter which is an array of the
158 blocknums of the secondary blocks which are to be produced. (Each element in
159 that array is required to be the blocknum of a secondary block, i.e. it is
160 required to be >= k and < m.)
162 The output from fec_encode() is the requested set of secondary blocks which are
163 written into output buffers provided by the caller.
165 Note that this fec_encode() is a "low-level" API in that it requires the input
166 data to be provided in a set of memory buffers of exactly the right sizes. If
167 you are starting instead with a single buffer containing all of the data then
168 please see easyfec.py's "class Encoder" as an example of how to split a single
169 large buffer into the appropriate set of input buffers for fec_encode(). If you
170 are starting with a file on disk, then please see filefec.py's
171 encode_file_stringy_easyfec() for an example of how to read the data from a file
172 and pass it to "class Encoder". The Python interface provides these
173 higher-level operations, as does the Haskell interface. If you implement
174 functions to do these higher-level tasks in other languages than Python or
175 Haskell, then please send a patch to zfec-dev@allmydata.org so that your API can
176 be included in future releases of zfec.
179 fec_decode() takes as input an array of k pointers, where each pointer points to
180 a buffer containing a block. There is also a separate input parameter which is
181 an array of blocknums, indicating the blocknum of each of the blocks which is
184 The output from fec_decode() is the set of primary blocks which were missing
185 from the input and had to be reconstructed. These reconstructed blocks are
186 written into output buffers provided by the caller.
191 encode() and decode() take as input a sequence of k buffers, where a "sequence"
192 is any object that implements the Python sequence protocol (such as a list or
193 tuple) and a "buffer" is any object that implements the Python buffer protocol
194 (such as a string or array). The contents that are required to be present in
195 these buffers are the same as for the C API.
197 encode() also takes a list of desired blocknums. Unlike the C API, the Python
198 API accepts blocknums of primary blocks as well as secondary blocks in its list
199 of desired blocknums. encode() returns a list of buffer objects which contain
200 the blocks requested. For each requested block which is a primary block, the
201 resulting list contains a reference to the apppropriate primary block from the
202 input list. For each requested block which is a secondary block, the list
203 contains a newly created string object containing that block.
205 decode() also takes a list of integers indicating the blocknums of the blocks
206 being passed int. decode() returns a list of buffer objects which contain all
207 of the primary blocks of the original data (in order). For each primary block
208 which was present in the input list, then the result list simply contains a
209 reference to the object that was passed in the input list. For each primary
210 block which was not present in the input, the result list contains a newly
211 created string object containing that primary block.
213 Beware of a "gotcha" that can result from the combination of mutable data and
214 the fact that the Python API returns references to inputs when possible.
216 Returning references to its inputs is efficient since it avoids making an
217 unnecessary copy of the data, but if the object which was passed as input is
218 mutable and if that object is mutated after the call to zfec returns, then the
219 result from zfec -- which is just a reference to that same object -- will also
220 be mutated. This subtlety is the price you pay for avoiding data copying. If
221 you don't want to have to worry about this then you can simply use immutable
222 objects (e.g. Python strings) to hold the data that you pass to zfec.
226 The Haskell code is fully Haddocked, to generate the documentation, run
227 % runhaskell Setup.lhs haddock
232 The filefec.py module has a utility function for efficiently reading a file and
233 encoding it piece by piece. This module is used by the "zfec" and "zunfec"
234 command-line tools from the bin/ directory.
239 A C compiler is required. To use the Python API or the command-line tools a
240 Python interpreter is also required. We have tested it with Python v2.4 and
241 v2.5. For the Haskell interface, GHC >= 6.8.1 is required.
246 Thanks to the author of the original fec lib, Luigi Rizzo, and the folks that
247 contributed to it: Phil Karn, Robert Morelos-Zaragoza, Hari Thirumoorthy, and
248 Dan Rubenstein. Thanks to the Mnet hackers who wrote an earlier Python wrapper,
249 especially Myers Carpenter and Hauke Johannknecht. Thanks to Brian Warner and
250 Amber O'Whielacronx for help with the API, documentation, debugging,
251 compression, and unit tests. Thanks to Adam Langley for improving the C API and
252 contributing the Haskell API. Thanks to the creators of GCC (starting with
253 Richard M. Stallman) and Valgrind (starting with Julian Seward) for a pair of
254 excellent tools. Thanks to my coworkers at Allmydata -- http://allmydata.com --
255 Fabrice Grinda, Peter Secor, Rob Kinninmont, Brian Warner, Zandr Milewski,
256 Justin Boreta, Mark Meras for sponsoring this work and releasing it under a Free