-
Notifications
You must be signed in to change notification settings - Fork 0
/
jerasure.h
284 lines (208 loc) · 11.5 KB
/
jerasure.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
/* jerasure.h - header of kernel procedures
* James S. Plank
JERASURE - Library for Erasure Coding
Copright (C) 2007 James S. Plank
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
James S. Plank
Department of Electrical Engineering and Computer Science
University of Tennessee
Knoxville, TN 37996
plank@cs.utk.edu
*/
/*
* $Revision: 1.2 $
* $Date: 2008/08/19 17:40:58 $
*/
#ifndef _JERASURE_H
#define _JERASURE_H
/* This uses procedures from the Galois Field arithmetic library */
#include "galois.h"
/* ------------------------------------------------------------ */
/* In all of the routines below:
k = Number of data devices
m = Number of coding devices
w = Word size
data_ptrs = An array of k pointers to data which is size bytes.
Size must be a multiple of sizeof(long).
Pointers must also be longword aligned.
coding_ptrs = An array of m pointers to coding data which is size bytes.
packetsize = The size of a coding block with bitmatrix coding.
When you code with a bitmatrix, you will use w packets
of size packetsize.
matrix = an array of k*m integers.
It represents an m by k matrix.
Element i,j is in matrix[i*k+j];
bitmatrix = an array of k*m*w*w integers.
It represents an mw by kw matrix.
Element i,j is in matrix[i*k*w+j];
erasures = an array of id's of erased devices.
Id's are integers between 0 and k+m-1.
Id's 0 to k-1 are id's of data devices.
Id's k to k+m-1 are id's of coding devices:
Coding device id = id-k.
If there are e erasures, erasures[e] = -1.
schedule = an array of schedule operations.
If there are m operations, then schedule[m][0] = -1.
operation = an array of 5 integers:
0 = operation: 0 for copy, 1 for xor (-1 for end)
1 = source device (0 - k+m-1)
2 = source packet (0 - w-1)
3 = destination device (0 - k+m-1)
4 = destination packet (0 - w-1)
*/
/* --------------------------------------------------------------- */
/* Bitmatrices / schedules ---------------------------------------- */
/*
- jerasure_matrix_to_bitmatrix turns a m X k matrix in GF(2^w) into a
wm X wk bitmatrix (in GF(2)). This is
explained in the Cauchy Reed-Solomon coding
paper.
- jerasure_dumb_bitmatrix_to_schedule turns a bitmatrix into a schedule
using the straightforward algorithm -- just
schedule the dot products defined by each
row of the matrix.
- jerasure_smart_bitmatrix_to_schedule turns a bitmatrix into a schedule,
but tries to use previous dot products to
calculate new ones. This is the optimization
explained in the original Liberation code paper.
- jerasure_generate_schedule_cache precalcalculate all the schedule for the
given distribution bitmatrix. M must equal 2.
- jerasure_free_schedule frees a schedule that was allocated with
jerasure_XXX_bitmatrix_to_schedule.
- jerasure_free_schedule_cache frees a schedule cache that was created with
jerasure_generate_schedule_cache.
*/
int *jerasure_matrix_to_bitmatrix(int k, int m, int w, int *matrix);
int **jerasure_dumb_bitmatrix_to_schedule(int k, int m, int w, int *bitmatrix);
int **jerasure_smart_bitmatrix_to_schedule(int k, int m, int w, int *bitmatrix);
int ***jerasure_generate_schedule_cache(int k, int m, int w, int *bitmatrix, int smart);
void jerasure_free_schedule(int **schedule);
void jerasure_free_schedule_cache(int k, int m, int ***cache);
/* ------------------------------------------------------------ */
/* Encoding - these are all straightforward. jerasure_matrix_encode only
works with w = 8|16|32. */
void jerasure_do_parity(int k, char **data_ptrs, char *parity_ptr, int size);
void jerasure_matrix_encode(int k, int m, int w, int *matrix,
char **data_ptrs, char **coding_ptrs, int size);
void jerasure_bitmatrix_encode(int k, int m, int w, int *bitmatrix,
char **data_ptrs, char **coding_ptrs, int size, int packetsize);
void jerasure_schedule_encode(int k, int m, int w, int **schedule,
char **data_ptrs, char **coding_ptrs, int size, int packetsize);
/* ------------------------------------------------------------ */
/* Decoding. -------------------------------------------------- */
/* These return integers, because the matrix may not be invertible.
The parameter row_k_ones should be set to 1 if row k of the matrix
(or rows kw to (k+1)w+1) of th distribution matrix are all ones
(or all identity matrices). Then you can improve the performance
of decoding when there is more than one failure, and the parity
device didn't fail. You do it by decoding all but one of the data
devices, and then decoding the last data device from the data devices
and the parity device.
jerasure_schedule_decode_lazy generates the schedule on the fly.
jerasure_matrix_decode only works when w = 8|16|32.
jerasure_make_decoding_matrix/bitmatrix make the k*k decoding matrix
(or wk*wk bitmatrix) by taking the rows corresponding to k
non-erased devices of the distribution matrix, and then
inverting that matrix.
You should already have allocated the decoding matrix and
dm_ids, which is a vector of k integers. These will be
filled in appropriately. dm_ids[i] is the id of element
i of the survivors vector. I.e. row i of the decoding matrix
times dm_ids equals data drive i.
Both of these routines take "erased" instead of "erasures".
Erased is a vector with k+m elements, which has 0 or 1 for
each device's id, according to whether the device is erased.
jerasure_erasures_to_erased allocates and returns erased from erasures.
*/
int jerasure_matrix_decode(int k, int m, int w,
int *matrix, int row_k_ones, int *erasures,
char **data_ptrs, char **coding_ptrs, int size);
int jerasure_bitmatrix_decode(int k, int m, int w,
int *bitmatrix, int row_k_ones, int *erasures,
char **data_ptrs, char **coding_ptrs, int size, int packetsize);
int jerasure_schedule_decode_lazy(int k, int m, int w, int *bitmatrix, int *erasures,
char **data_ptrs, char **coding_ptrs, int size, int packetsize,
int smart);
int jerasure_schedule_decode_cache(int k, int m, int w, int ***scache, int *erasures,
char **data_ptrs, char **coding_ptrs, int size, int packetsize);
int jerasure_make_decoding_matrix(int k, int m, int w, int *matrix, int *erased,
int *decoding_matrix, int *dm_ids);
int jerasure_make_decoding_bitmatrix(int k, int m, int w, int *matrix, int *erased,
int *decoding_matrix, int *dm_ids);
int *jerasure_erasures_to_erased(int k, int m, int *erasures);
/* ------------------------------------------------------------ */
/* These perform dot products and schedules. -------------------*/
/*
src_ids is a matrix of k id's (0 - k-1 for data devices, k - k+m-1
for coding devices) that identify the source devices. Dest_id is
the id of the destination device.
jerasure_matrix_dotprod only works when w = 8|16|32.
jerasure_do_scheduled_operations executes the schedule on w*packetsize worth of
bytes from each device. ptrs is an array of pointers which should have as many
elements as the highest referenced device in the schedule.
*/
void jerasure_matrix_dotprod(int k, int w, int *matrix_row,
int *src_ids, int dest_id,
char **data_ptrs, char **coding_ptrs, int size);
void jerasure_bitmatrix_dotprod(int k, int w, int *bitmatrix_row,
int *src_ids, int dest_id,
char **data_ptrs, char **coding_ptrs, int size, int packetsize);
void jerasure_do_scheduled_operations(char **ptrs, int **schedule, int packetsize);
/* ------------------------------------------------------------ */
/* Matrix Inversion ------------------------------------------- */
/*
The two matrix inversion functions work on rows*rows matrices of
ints. If a bitmatrix, then each int will just be zero or one.
Otherwise, they will be elements of gf(2^w). Obviously, you can
do bit matrices with crs_invert_matrix() and set w = 1, but
crs_invert_bitmatrix will be more efficient.
The two invertible functions return whether a matrix is invertible.
They are more efficient than the inverstion functions.
Mat will be destroyed when the matrix inversion or invertible
testing is done. Sorry.
Inv must be allocated by the caller.
The two invert_matrix functions return 0 on success, and -1 if the
matrix is uninvertible.
The two invertible function simply return whether the matrix is
invertible. (0 or 1). Mat will be destroyed.
*/
int jerasure_invert_matrix(int *mat, int *inv, int rows, int w);
int jerasure_invert_bitmatrix(int *mat, int *inv, int rows);
int jerasure_invertible_matrix(int *mat, int rows, int w);
int jerasure_invertible_bitmatrix(int *mat, int rows);
/* ------------------------------------------------------------ */
/* Basic matrix operations -------------------------------------*/
/*
Each of the print_matrix routines require a w. In jerasure_print_matrix,
this is to calculate the field width. In jerasure_print_bitmatrix, it is
to put spaces between the bits.
jerasure_matrix_multiply is a simple matrix multiplier in GF(2^w). It returns a r1*c2
matrix, which is the product of the two input matrices. It allocates
the product. Obviously, c1 should equal r2. However, this is not
validated by the procedure.
*/
void jerasure_print_matrix(int *matrix, int rows, int cols, int w);
void jerasure_print_bitmatrix(int *matrix, int rows, int cols, int w);
int *jerasure_matrix_multiply(int *m1, int *m2, int r1, int c1, int r2, int c2, int w);
/* ------------------------------------------------------------ */
/* Stats ------------------------------------------------------ */
/*
jerasure_get_stats fills in a vector of three doubles:
fill_in[0] is the number of bytes that have been XOR'd
fill_in[1] is the number of bytes that have been copied
fill_in[2] is the number of bytes that have been multiplied
by a constant in GF(2^w)
When jerasure_get_stats() is called, it resets its values.
*/
void jerasure_get_stats(double *fill_in);
#endif