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
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
|
<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<link href="style.css" rel="stylesheet">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta content="text/html; charset=ISO-8859-1" http-equiv="content-type">
<title>Benchmarking backup tools</title>
</head>
<body>
<div class=top>
<img alt="DAR's Documentation" src="dar_s_doc.jpg" style="float:left;">
<h1>Benchmarking backup tools</h1>
</div>
<div class=main>
<h2>Introduction</h2>
<p>
This document has for objective to compare common backup tools under Unix
(Linux, FreeBSD, MACOS X...), among the most commonly available today.
</p>
<ul>
<li>
The <b>first target</b> we want to address is being able to copy a directory tree and files
with the best fidelity,
</li>
<li>
The <b>second target</b> is being able to backup and restore a whole system
from a minimal environment without assistance of an already existing local server (disaster context).
</li>
<li>
The <b>third target</b> is being able to securely keep for the long term an archived data. Securely here means having the ability to detect data corruption and limit its impact on the rest of the archive.
</li>
</ul>
<p>
Depending on the targets we may need compression and/or ciphering inside backup,
but also denpending on the context (public cloud storage, removable media, ...), limited storage space.
</p>
<p>
Backup softwares that requires servers already running
on the local network (For examples <i>Bacula</i>, <i>Amanda</i>, <i>Bareos</i>, <i>UrBackup</i>, <i>Burp</i>...) cannot
address our second target as we would have first to reconstruct such server in
case of disaster (from what then?) in order be able to restore our system and its data. They are
over complex for the first target and are not suitable for the third.
</p>
<p>
Partition cloning systems (<i>clonezilla</i>, <i>MondoRescue</i>, <i>RescueZilla</i>,
<i>partclone</i>, <i>dump</i> and consorts) are targetted at block copy and as such cannot backup a live system:
you have to shutdown and boot on a CD/USB key or run in single user-mode in order to "backup". This cannot
be automated and has a strong impact on the user as she/he has to interrupt her/his work during the whole
backup operation.
<p>
Looking at the remaining backup tools, with or without Graphical User Interface, most of them
rely on one of the three backend softwares, <i>tar</i>, <i>rsync</i> and <i>dar</i>:
</p>
<ul>
<li>Software based on <b>dar</b>: gdar, DarGUI, Baras, Darbup, Darbrrd, HUbackup, SaraB...</li>
<li>Software based on <b>rsync</b>: TimeShift, rsnapshot... </li>
<li>Software based on <b>tar</b>: BackupPC, Duplicity, fwbackups... </li>
</ul>
<p>
We will thus compare these three softwares for the different test famillies described below.
</p>
<h2>Tests Famillies</h2>
<p>
Several aspects are to be considered:
</p>
<lu>
<li><b>completness</b> of the restoration: file permissions, dates precision, hardlinks, file attributes, Extended Attributes, sparse files...</li>
<li><b>main features</b> around backup: differential backup, snapshot, deduplication, compression, encrytion, file's history...</li>
<li><b>robustness</b> of the backup: how data corruption impact the backup, how it is reported...</li>
<li><b>execution performance</b>: execution time, memory consumption, multi-threading support...</li>
</lu>
<h2>Benchmark Results</h2>
<p>
The results presented here are a synthesis of the <a href="benchmark_logs.html">test logs</a>. This synthesis
is in turn summarized one step further in conclusion of this document.
</p>
<h3>Completness of backup and restoration</h3>
<div class=table>
<table class=center>
<tr>
<th>Software</th>
<th>plain file</th>
<th>symlink</th>
<th>hardlinked files</th>
<th>hardlinked sockets</th>
<th>hardlinked pipes</th>
<th>user</th>
<th>group</th>
<th>perm.</th>
<th>ACL</th>
<th>Extended Attributes</th>
<th>FS Attributes</th>
<th>atime</th>
<th>mtime</th>
<th>ctime</th>
<th>btime</th>
<th>Spares File</th>
<th>Disk usage optimization</th>
</tr>
<tr>
<td>Dar</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>-</td>
<td>yes(1)</td>
<td>yes</td>
<td>yes</td>
</tr>
<tr>
<td>Rsync</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes(4)</td>
<td>yes(5)</td>
<td>-</td>
<td>-</td>
<td>yes</td>
<td>-</td>
<td>yes(1)</td>
<td>yes(6)</td>
<td>yes(6)</td>
</tr>
<tr>
<td>Tar</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>- <i>(2)</i></td>
<td>-</td>
<td>yes</td>
<td>yes</td>
<td>yes</td>
<td>yes(7)</td>
<td>yes(8)</td>
<td>-</td>
<td>-</td>
<td>yes(3)</td>
<td>-</td>
<td>yes(1)</td>
<td>yes(6)</td>
<td>-</td>
</tr>
</table>
</div>
<ul>
<li>(1) "Yes" under MACoS X, FreeBSD and BSD systems. As of today (year 2020), Linux has no way to set the <i>btime</i> aka <i>birthtime</i> or yet <i>creation time</i></li>
<li>(2) <i>tar</i> does even not save and restore plain normal sockets, but that's not a big issue in fact as Unix sockets should be recreated by the applications that provide the corresponding service</li>
<li>(3) unless <code>--xattrs</code> is provided, <i>mtime</i> is saved by <i>tar</i> but with an accuracy of only 1 second, while today's systems provide nanosecond precision</li>
<li>(4) needs -A option</li>
<li>(5) needs -X option</li>
<li>(6) needs -S option</li>
<li>(7) needs --acl option</li>
<li>(8) needs --xattrs option</li>
</ul>
</p>
<p>
See <a href="benchmark_logs.html#completness">the test logs</a> for all the details.
</p>
<h3>Feature set</h3>
<p>
In addition to the exhaustivity of the restored data (seen above), several features are a
<i>must have</i> when creating backups. Their description and what they bring to a backup process
is given below, followed by a table of how they are supported on the different softwares under test:
</p>
<dl>
<dt>Historization</dt><dd>
Historization is the ability to restore a deleted file even long after the mistake has been made by
rotating backups over an arbitrary large number of backup set. Having associated tools
to quickly locate the backup where resides a particular file's version becomes important
when the history increases. Historization can be done with only full backups, but of course better
leverages differential and incremental backups.
</dd>
<br/>
<dt>Data filtering</dt><dd>
Not all files need to be saved:
<ul>
<li>
some <b>directories</b> (like <code>/tmp, /proc, /sys, /dev, /home/*/.cache</code>) are useless to save
</li>
<li>
some <b>files</b> based on their name or part of their name --- their extension for
example, (like emacs's backup files <code>*~</code> or your music files<code>*.mp3</code> you already
have archives somewhere, and so on) need not to be saved neither.
</li>
<li>
You may wish to ignore files located one or more particular <b>mounted filesystem</b>, or at the
opposite, only consider certains volume/disk/mounted filesystem and ignore all others, and have different
backup rotation cycles for those.
</li>
<li>
You may also find better to <b>tag</b> files one by one (manually by mean of an automated process of
your own), to be excluded from or included in the backup
</li>
<li>
Instead of tagging you could also let a process define a long <b>file listing</b> to backup and/or to ignore.
</li>
<li>
Last, you may well need a mix of several of these mechanisms at the same time
</li>
</ul>
</dd>
<br/>
<dt>Slicing (or multi-volume)</dt><dd>
Having a backup split into several files of given max size can address several needs:
<ul>
<li>hold the backup on several removal media (CD, DVD, USB keys...) smaller than the backup itself</li>
<li>transfer the backup from a large space to another by mean of a smaller removable media</li>
<li>transfer the backup over the network and recover at the last transmitted slice rather than restarting the whole
transfer in case of network issue</li>
<li>store the backup int the cloud where the provider limits the file size</li>
<li>be able to restore a backup on a system where storage space cannot hold both the backup and the restored system</li>
<li>transfer back from the cloud only a few slices to restore some files, when cloud provider does not provide adhoc protocols (sftp, ftp, ...) but only a user web based interface</li>
</ul>
Of course, multi-volume is really interesting if you don't have to concatenate all the slices to be able to have a usable backup.
<br/>
<br/>
Last the previously identified use cases for backup slicing turn around limited storage space, thus having compression available when
multi-volume is used is a key point here.
</dd>
<br/>
<dt>Symmetric strong encryption</dt><dd>
Symmetric strong encryption is the ability to cipher a backup with a password or passphrase and use that same key to decipher it. Some
well known algorithms in this area are AES, blowfish, camellia...
<br/>
Symmetric strong encryption is interesting for the following cases:
<ul>
<li>if your disk is ciphered, would you store your backup in clear on the cloud?</li>
<li>you do not trust your cloud provider to not inspect your data and make marketing profile of yourself with it.</li>
<li>You want to prevent your patented data or industrial secret recipies from falling into the competition's hands or goverment agencies that could clone it without fear of being prosecuted. This use case applies whether your backup is stored on local disk, removable media or public cloud.</li>
<li>Simply because in your country, you have the right and the freedom to have privacy.</li>
<li>Because your today democratic country could tomorrow verse into a dictatorship and based on some arbitrary criteria,
(belief, political opinion, sexual orientation...) you could suffer tomorrow from having this information having been accessible
today to the authorities or even having been publicly released, while you still need backup using arbitrary storage medium.
</li>
</ul>
</dd>
<br/>
<dt>Asymmetric strong encryption</dt><dd>
Asymmetrical strong encryption is the ability to cipher a backup with a public key and having the corresponding private key for deciphering it (PGP, GnuPG...).
<br/>
Asymmetric encrypion is mainly interesting when exchanging data over Internet between different persons, or eventually for archiving data in the public cloud.
Having it for backup seems not appropriate and is more complex than symmetric strong encryption, as restoration requires
the private key, which thus must be stored outside the backup itself still be protected from unauthorized access.
The private key use can still be protected with a password or a passphrase
but this gives the same feature level as symmetrical encryption with a more complex process and not much more security.
</dd>
<br/>
<dt>Protection against plain-text attack</dt>
<dd>Ciphering data must be done with a minimum level of security, in particular when the ciphered data has well defined
structure and patterns, like a backup file format is expected to have. Knowing such expected structure of the clear data
may lead an attacker to undisclose the whole ciphered data. This is known as <i>plain-text attack</i>.
</dd>
<br/>
<dt>Key derivation function</dt><dd>
<ul>
<li>
Using the same password/passphrase for different backups is convenient but not secure. Having a key derivation function
using a <i>salt</i> let you use the same password/passphrase while the data will be encrypted with a different key each time,
this is the role of the <i>Key Derivation Function (KDF)</i> (PKCS5/PBKDF2, Argon2...).
</li>
<li>
Another need for a KDF is that usually the human provided
password/passphrase are weak: Even when we use letters, digits and some special characters, passwords and passphrases are still located in a
small area of possible keys that a <i>dictionnary attack</i> can leverage. As the KDF is also by design CPU intensive,
it costs a lot of effort and time to an attacker to derive each word of a dictionnary to its resulting KDF transformed words.
The required time to perform a dictionnary attack can thus be multiplied by several hundred thousand times,
leading to an effective time of tens of years and even centuries rather than hours or days.
</li>
</ul>
</dd>
<br/>
<dt>File change detection</dt><dd>
When backing up a live system, it is important to detect, retry saving or flag files that changed during the time
they were read for backup. In such situation, the backed file could be recorded in a state it never had: As the backup process
reads sequentially from the beginning to the end, if a modification <i>A</i> is done at the end of file then a
modification <i>B</i> is made at its beginning during this file's backup, the backup may contain <i>B</i> and not <i>A</i>
while at not time
the file contained <i>B</i> without <i>A</i>. Seen the short time a file can be read, time accuracy of micro or nanoseconds
is mandatory to detect such file change during a backup process, else you will screw up your data in the backup and have nothing
to rely on in the occurence of a deleted file by mistake, disk crash or disaster.
<br/>
At restoration time, if the file has been saved anyway, it should be good to know the such file was not saved properly, maybe
restoring a older version but a sane one would be better. Something the user/sysadmin cannot guess if the backup does not hold
such type of information.
</dd>
<br/>
<dt>Multi-level backup</dt><dd>
Multi-level backup is the ability to make use of <b>full</b> backups, <b>differential</b> backups and/or eventually <b>incremental</b> backups.
<br/>
The advantage of differential and incremental backups compared to full ones is the much shorter time they require to complete
and the reduces storage space and/or bandwidth they imply when transfered over the network.
</dd>
<br/>
<dt>Binary delta</dt>
<dd>Without binary delta, when performing a differential or incremental backup, if a file has changed since the previous
backup, it will be resaved entirely. Some huge files made by some well know applications (mailboxes for example) would consume
a lot of storage space and lead to a long backup time even when performing incremental or differential backups. Binary delta is
the ability to only store the part of a file that changed since a reference state, this lead to important space gain and reduction
of the backup duration.
</dd>
<br/>
<dt>Detecting suspicious modifications</dt>
<dd>When performing a backup based on a previous one (differential, incremental, decremental backups), it is possible
to check the way the metadata of saved files have changed until then and warn the user when some uncommon pattens are met.
Those may be the trace of a rootkit, virus, ransomware or trojan, trying to hide its presence and activities.
</dd>
<br/>
<dt>Snapshot</dt><dd>
A snapshot is like a differential backup made right after the full backup (no file has changed): it is a minimal
set of information that can be used to:
<ul>
<li>
create an incremental or differential backup without having the full backup around
or more generally the backup of reference: When backup are stored remotely, snapshot is a must.
</li>
<li>
compare the current living filesystem with a status it had at the time the snapshot was made
</li>
<li>
bring some metadata redundancy and repairing mean to face a corrupted backup
</li>
</ul>
</dd>
<br/>
<dt>On-fly hashing</dt><dd>
On-fly hashing is the ability to generate a hashing of the backup at the same time it is generated and before it is written
to storage. Such hash can be used to:
<ul>
<li>validate a backup has been properly transfered to a public storage cloud having hash computation done in parallel</li>
<li>check that no data corruption has occured (doubt about disk or memory) even when the backup is written to local disk</li>
</ul>
Hashing validation is usually faster than backup testing or backup comparison, though it does not validate your ability
to rely on the backup as deeply as these later operations. Hashing can be made after the backup has been completed but
it will need to re-read the whole backup and you will have to wait for the necessary storage I/O for the operation to complete.
On-fly hashing should leverage the fact the data is in memory so it saves the corresponding disk I/O and corresponding
latency, thus it is much faster. As it is also done in memory it can help detect file corruption on the backup destination media
(like USB keys or poor quality hardware).
</dd>
<br/>
<dt>Run custom command during operation</dt><dd>
For an automated backup process, it is often necessary to run commands before and after the backup operation itself.
But also during the backup process. For example, when entering a directory, one could need to run an arbitrary command
generating a file that will be included in the backup. Or while exiting such directory performing some cleanup operation in that same directory.
Another use case is found when slicing the backup, by the ability to perform after each slice generation a custom operation like uploading the
slice to cloud, burning to DVD-/+RW, loading a tape from a tape library...
</dd>
<br/>
<dt>Dry-run execution</dt><dd>
When tuning a backup process, it is often necessary to verify quickly that all will work flawlessly without having
to wait for a backup to complete, consume storage resource and network bandwidth.
</dd>
<br/>
<dt>User message within backup</dt><dd>
Allowing the user to add an arbitrary message within the backup may be useful when the filename is too small
to hold the needed information (like the context the backup or archive was made, hint for the passphrase... and so on).
</dd>
<br/>
<dt>Backup sanity test</dt><dd>
It is crutial in a backup process to validate that the generated
backup is usable. There are many reasons it could not be the case, from
a data corruption in memory, on disk or over the network ; a disk space saturation
leading to truncated backup, down to a software bug.
</dd>
<br/>
<dt>Comparing with original data</dt><dd>
One step further for backup and archiving validation is compairing file content and metadata with the system it has.
</dd>
<br/>
<dt>Tunable verbosity</dt><dd>
When a backup process is in production and works nicely, it is usually interesting to have the minimal output possible
for that any error still be possible to log. While when setting up a backup process, having more detailed
information is required to understand and validate that the backup process follows the expected path.
</dd>
<br/>
<dt>Modify the backup's content</dt><dd>
Once a backup has been completed, you might notice that you have saved extra files you ought not to save. Being able to drop
them from the backup to save some space without having to restart the whole backup may lead to a huge time saving.
<br/>
<br/>
You might also need to add some extra files that were outside the backup scope, having the possibility to add them
without restarting the whole backup process may also lead to a huge time saving.
</dd>
<br/>
<dt>Stdin/stdout backup read/write</dt><dd>
Having the ability to pipe the generated backup to an arbitrary command is on of the ultimate key of
backup software flexibility.
</dd>
<br/>
<dt>Remote network storage</dt><dd>
This is the ability to produce directly a backup to a network storage without using local disk, and to
be able to restore directly reading a backup from the such remote storage still without using local storage.
<i>Network/Remote storage</i> is to be understood as remote network storage like public cloud, private cloud,
personal NAS... that are accesible from the network by mean of a file transfer protocols (scp, sftp, ftp,
rcp, http, https...)
</dd>
<br/>
</dl>
<div class=table>
<table class=center>
<tr class=center>
<th width="40%">Feature</th>
<th width="20%">Dar</th>
<th width="20%">Rsync</th>
<th width="20%">Tar</th>
<tr>
<tr>
<th class=left>Historization</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Data filtering by directory</th>
<td>Yes</td>
<td>Yes</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Data filtering by filename</th>
<td>Yes</td>
<td>Yes</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Data filtering by filesystem</th>
<td>Yes</td>
<td>limited</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Data filtering by tag</th>
<td>limited</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>Data filtering by files listing</th>
<td>Yes</td>
<td>yes</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Slicing/multi-volume</th>
<td>Yes</td>
<td>-</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Symmetric encryption</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Asymmetric encryption</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Plain-text attack protection</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>PBKDF2 Key Derivation Function</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>ARGON2 Key Derivation Function</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>File change detection</th>
<td>Yes</td>
<td>-</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Multi-level backup</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Binary delta</th>
<td>Yes</td>
<td>Yes</td>
<td>-</td>
</tr>
<tr>
<th class=left>Detecting suspicious modifications</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>Snapshot for diff/incr. backup</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Snapshot for comparing</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>Snapshot for redundancy</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>On-fly hashing</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>Run custom command during operation</th>
<td>Yes</td>
<td>-</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Dry-run execution</th>
<td>Yes</td>
<td>Yes</td>
<td>-</td>
</tr>
<tr>
<th class=left>User message within backup</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<th class=left>Backup sanity test</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Comparing with original data</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Tunable verbosity</th>
<td>Yes</td>
<td>Yes</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Modify the backup's content</th>
<td>Yes</td>
<td>Yes</td>
<td>limited</td>
</tr>
<tr>
<th class=left>Stdin/stdout backup read/write</th>
<td>Yes</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Remote network storage</th>
<td>Yes</td>
<td>limited</td>
<td>Yes</td>
</tr>
</table>
</div>
<p>
The presented results above is a synthesis of the <a href="benchmark_logs.html#features">test logs</a>
</p>
<h3>Robustness</h3>
<p>
The objective here is to see how a minor data corruption can impacts the backup. Such type of
corruption (a single bit invertion) can be caused by network transfert, cosmic particle hitting
the memory bank, or simply due to the time passing stored on a particular medium. In real life
data corruption may impact more than one bit, right. But if the ability to workaround the corruption of a
single bit does not bring any information about the ability to recover larger volume
of data corruption, the <u>inability</u> to recover a single bit, is enough to know that the same software
will behave even worse when larger portion of data corruption will be met.
</p>
<div class=table>
<table class=center>
<tr>
<th style="min-width:30%">Behavior</th>
<th style="min-width:10%">Dar</th>
<th style="min-width:10%">Rsync</th>
<th style="min-width:10%">Tar alone</th>
<th style="min-width:10%">Tar + gzip</th>
</tr>
<tr>
<th class=left>Detects backup corruption</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
<td>Yes</td>
</tr>
<tr>
<th class=left>Warn or avoid restoring corrupted data</th>
<td>Yes</td>
<td>-</td>
<td>-</td>
<td>Yes</td>
<tr>
<tr>
<th class=left>Able to restore all files not concerned by the corruption</th>
<td>Yes</td>
<td>Yes</td>
<td>Yes</td>
<td>-</td>
</tr>
</table>
</div>
<p>
To protect your data, you can go one step further computing data redundancy with <a href="https://github.com/Parchive/par2cmdline">Parchive</a>
on top of your backup or archives. This will allow you to repair them in case of corruption.
<ul>
<li>
Though, <i>rsync</i> is not adapted to that process as creating a global redundancy of a directory tree is much more complex and error-prone.
At the opposite, <i>tar</i> and <i>dar</i> are pretty well adapted as a backup may be
a single file or a few big files if using slicing or multi-volume backup.
</li>
<li>
Second, whatever is the redundancy level you select, if the data corruption exceed this level, you will not be able to repair your backups and
archives. Thus, better relying on a robust and redundant backup file structure, and here <i>dar</i> has some big advantages.
</li>
<li>
Last, if execution time is important for you, having a sliced backup with a slice size smaller than the available RAM
and running <i>Parchive</i> right after each slice created, will save a lot of disk I/O and can <b>speed up the overall process
by more than 40%</b>. But here too, only <i>dar</i> provides this possibility.
</li>
</ul>
</p>
<p>
The presented results above is a synthesis of the <a href="benchmark_logs.html#robustness">test logs</a>.
</p>
<h3>Performance</h3>
<p>
In the following, we have distinguished two purposes of backup tools: the "identical" copy of a set of files and directories (short term operation) and the
usual backup operation (long term storage and historization).
</p>
<h4>Performance of file copy operation</h4>
<p>
The performance aspect to consider for this target is exclusively the execution speed, this may imply data reduction
on the wire only if the bandwidth is low enough for the compression processing time added does not ruine the gain on
transfer time. Compression time is not dependent on the backup tool but on the data, and we will see in the backup
performances tests, the way the different backup tools do reduce data on the wire. For the execution time we get the following
results:
</p>
<h5>Single huge file</h5>
<p>The copied data was a Linux distro installation ISO file</p>
<div class="cadre">
<div class="gauge best" style="width: 17%;">cp: 2.58 s</div>
<div class="gauge normal" style="width: 61%;">Dar: 9.18 s</div>
<div class="gauge normal" style="width: 100%">Rsync: 15.28 s</div>
<div class="gauge normal" style="width: 43%">Tar: 6.51 s</div>
</div>
<h5>Linux system</h5>
<p>The copied data was a fresh fully featured Linux installed system</p>
<div class="cadre">
<div class="gauge best" style="width: 31%;">cp: 5.15 s</div>
<div class="gauge normal" style="width: 100%;">Dar: 16.78 s</div>
<div class="gauge normal" style="width: 99%">Rsync: 16.59 s</div>
<div class="gauge normal" style="width: 48%">Tar: 8.04 s</div>
</div>
<h5>Conclusion</h5>
<p>
for local copy <b><i>cp</i></b> is the fastest but totally unusable for remote copy. At first sight one could think
<i>tar</i> would be the best alternative for remote copy, but that would not take into account the fact you will probably want
to use secured connection (unless all segments of the underlying network are physically yours, end to end). Thus once the
backup will be generated, using <i>tar</i> will require an extra user operation, extra computing time to cipher/decipher and time to
transfer the data while both alternatives, <i>rsync</i> and <i>dar</i>, have it integrated: they can copy and transfer at the
same time, with both the gain of time and the absence of added operations for the user.
</p>
<p>
In consequence, for remote copy, if this is for a unique/single remote copy, <b><i>dar</i></b> will be faster than <i>rsync</i> most of the time
(even when using compression to cope with low bandwidth, see the backup test results, below). But for recurring remote copy even if <b><i>rsync</i></b> is not faster that
<i>dar</i>, it has the advantage of being designed espetially for this task as in that context we do not need to store the data compressed nor ciphered.
Things we can summarize as follows:
</p>
<div class=table>
<table class=center>
<tr>
<th>Operation</th>
<th>Best Choice</th>
<th>Alternative</th>
</tr>
<tr>
<th class=left>Local copy</th>
<td>cp</td>
<td>tar</td>
</tr>
<tr>
<th class=left>One-time remote copy</th>
<td>dar</td>
<td>rsync</td>
</tr>
<tr>
<th class=left>recurrent remote copy</th>
<td>rsync</td>
<td>dar</td>
</tr>
</table>
</div>
<p>
See the corresponding <a href="benchmark_logs.html#copy_perf">test logs</a> for more details
</p>
<h4>Performance of <a name="backup_perf">backup</a> operation</h4>
<p>
For backup we consider the following criteria by order of importance:
</p>
<ol>
<li>data reduction on backup storage</li>
<li>data reduction when transmitted over the network</li>
<li>execution time to restore a few files</li>
<li>execution time to restore a full and differential backups</li>
<li>execution time to create a full and differential backups</li>
</ol>
<p>
Why this order?
</p>
<ul>
<li>
Because usually backup creation is done at low priority in background and
on a day to day basis, the execution time is less important than reducing the storage usage: reducing storage usage gives longer
backup history and increases the ability
to recover accidentically removed files much later after the mistake has been done (which may be detected
weeks or months afterward).
</li>
<li>
Next, while your backup storage can be anything,
including low cost or high end dedicated one, we see more and more frequently externalized backups, which main declinaison
is based on public cloud storage, leading to relatively cheap disaster recovery solution. However, your WAN/Internet acces will
be drained by the backup volumes flying away and you probably don't want them to consume too much of this bandwidth which could
slow down your business or Internet access. As a workaround, one could rate-limit the bandwidth for backup exchanges only. But doing
so will extend the backup transfer time so much that you may have to reduce the backup frequency to not have two backups
transfered at the same time. This would lead you to lose accuracy of saved data: A too low backup frequency will only allow you
to restore your systems in the state they had several days instead of several hours or several tens of minutes, before the disaster
occured. For that reason data reduction on the wire is the second criterium. Note that data reduction on storage
usually implies data reduction on the wire, but the opposite is not always true, depending on the backup tool used.
</li>
<li>
Next, it is much more frequent to have
to restore a few files (corrupted or deleted by mistake) and we need this to be quick because this is an interactive operation and
that the missing data is mandatory to go forward for one's work, which workflow may impact several other persons.
</li>
<li>
The least frequent operation (hopefully) is the restoration of a whole system
in case of disaster. Having it performing quick is of course important, but less than having a complete, robust, accurate and
recent backup somewhere, that you can count on to restore your systems in the most recent possible state.
</li>
</ul>
<p>
Note that the following result do not take into account the performance penalty implied by the
<b>network latency</b>. Several reasons to that:
</p>
<ul>
<li>
it would not measure the software performance but the network bandwidth and latency which is not the object
of this benchmark and may vary with distance, link layer technology and number of devices crossed,
</li>
<li>
We can assume the network penalty to be proportional to data processed by each software, as all protocol used are usually TCP based
(ftp, sftp, scp, ssh, ...), which performance is related to the operating system parameters (window size, MTU, etc.) not to the backup software
itself. As we only rely on tmpfs
filesystems for this benchmark to avoid mesuring the disk I/O performance, we may approximate that a network latency increase or a reduction of network bandwidth would
just inflate the relative execution time of the different tested softwares in a linear manner. In other words, adding network between
system and backup storage should thus not modify the relative performances of the softwares under test.
</li>
</ul>
<p>
For all the backup performance tests that follow (but not for file copy performance tests seen above),
compression has been activated using the same and most commonly
supported algorithm: gzip at level 6. Other algorithms may complete faster or provide better compression ratio, but this is linked to
chosen compression algorithm and data to compress, not to the backup tools tested here.
</p>
<h4>Data reduction on backup storage</h4>
<h5>Full backup</h5>
<div class="cadre">
<div class="gauge normal" style="width: 38%;">Dar: 1580562224 bytes</div>
<div class="gauge normal" style="width: 38%">Dar+sparse: 1578428790 bytes</div>
<div class="gauge normal" style="width: 39%">Dar+sparse+binary delta: 1602481058 bytes</div>
<div class="gauge normal" style="width: 100%">Rsync: 4136318307 bytes</div>
<div class="gauge normal" style="width: 100%">Rsync+sparse: 4136318307 bytes</div>
<div class="gauge normal" style="width: 37%">tar: 1549799048 bytes</div>
<div class="gauge best" style="width: 37%">tar+sparse: 1549577862 bytes</div>
</div>
<h5>Differential backup</h5>
<div class="cadre">
<div class="gauge normal" style="width: 100%;">Dar: 49498524 bytes</div>
<div class="gauge normal" style="width: 100%">Dar+sparse: 49505251 bytes</div>
<div class="gauge best" style="width: 48%">Dar+sparse+binary delta: 23883368 bytes</div>
<div class="gauge ref" style="width: 100%">Rsync: not supported</div>
<div class="gauge ref" style="width: 100%">Rsync+sparse: not supported</div>
<div class="gauge normal" style="width: 90%">tar: 44607904 bytes</div>
<div class="gauge normal" style="width: 90%">tar+sparse: 44604194 bytes</div>
</div>
<h5>Full + Differential backup</h5>
<p>
This is a extrapolation of the required volume for backup, after one week of daily backup of the Linux system
under test, assuming the activity is as minimal each day as it was here between the initial day of the
full backup and the day of the first differential backup (a few package upgrade and no user activity).
</p>
<div class="cadre">
<div class="gauge normal" style="width: 100%;">Dar: 1927051892 bytes</div>
<div class="gauge normal" style="width: 100%">Dar+sparse: 1924965547 bytes</div>
<div class="gauge best" style="width: 92%">Dar+sparse+binary delta: 1769664634 bytes</div>
<div class="gauge ref" style="width: 100%">Rsync: not supported</div>
<div class="gauge ref" style="width: 100%">Rsync+sparse: not supported</div>
<div class="gauge normal" style="width: 97%">tar: 1862054376 bytes</div>
<div class="gauge normal" style="width: 97%">tar+sparse: 1861807220 bytes</div>
</div>
<p>
This previous results concerns the backup of a steady Linux system, relative difference of data reduction might favorize both <i>rsync</i>
and <i>dar+binary delta</i> when the proportion of large files being slightly modified increases (like mailboxe files).
</p>
<h4>Data reduction over network</h4>
<h5>Full backup</h5>
<div class="cadre">
<div class="gauge normal" style="width: 99%;">Dar: 1580562224 bytes</div>
<div class="gauge normal" style="width: 98%">Dar+sparse: 1578428790 bytes</div>
<div class="gauge normal" style="width: 100%">Dar+sparse+binary delta: 1602481058 bytes</div>
<div class="gauge normal" style="width: 99%">Rsync: 1587714486 bytes</div>
<div class="gauge normal" style="width: 99%">Rsync+sparse: 1587714474 bytes</div>
<div class="gauge normal" style="width: 97%">tar: 1549799048 bytes</div>
<div class="gauge best" style="width: 97%">tar+sparse: 1549577862 bytes</div>
</div>
<h5>Differential backup</h5>
<div class="cadre">
<div class="gauge normal" style="width: 100%;">Dar: 49498524 bytes</div>
<div class="gauge normal" style="width: 100%">Dar+sparse: 49505251 bytes</div>
<div class="gauge best" style="width: 48%">Dar+sparse+binary delta: 23883368 bytes</div>
<div class="gauge normal" style="width: 59%">Rsync: 29293958 bytes</div>
<div class="gauge normal" style="width: 59%">Rsync+sparse: 29293958 bytes</div>
<div class="gauge normal" style="width: 90%">tar: 44607904 bytes</div>
<div class="gauge normal" style="width: 90%">tar+sparse: 44604194 bytes</div>
</div>
<h5>Full + Differential backup</h5>
<p>
This is the same extrapolation done above (one week of daily backup), but for the volume of data transmitted over the network instead of the backup volume on storage.
</p>
<div class="cadre">
<div class="gauge normal" style="width: 100%;">Dar: 1927051892 bytes</div>
<div class="gauge normal" style="width: 100%">Dar+sparse: 1924965547 bytes</div>
<div class="gauge best" style="width: 92%">Dar+sparse+binary delta: 1769664634 bytes</div>
<div class="gauge normal" style="width: 93%">Rsync: 1792772192 bytes</div>
<div class="gauge normal" style="width: 93%">Rsync+sparse: 1792772180 bytes</div>
<div class="gauge normal" style="width: 97%">tar: 1862054376 bytes</div>
<div class="gauge normal" style="width: 97%">tar+sparse: 1861807220 bytes</div>
</div>
<h4>Execution time to restore a few files</h4>
<div class="cadre">
<div class="gauge normal" style="width: 3.9%;">Dar: 0.98 s</div>
<div class="gauge normal" style="width: 4.49%">Dar+sparse: 1.13 s</div>
<div class="gauge normal" style="width: 5.05%">Dar+sparse+binary delta: 1.27 s</div>
<div class="gauge best" style="width: 0.01%">Rsync: 3 ms</div>
<div class="gauge best" style="width: 1%">Rsync+sparse: 3 ms</div>
<div class="gauge normal" style="width: 100%">tar: 25.15 s </div>
<div class="gauge normal" style="width: 99%">tar+sparse: 25 s</div>
</div>
<p>
Here the phenomenum is even more important when the file to restore is located near the end of the <i>tar</i> backup,
as <i>tar</i> sequentially reads the whole backup up to the requested file.
</p>
<h4>Execution time to restore a whole system - full backup</h4>
<div class="cadre">
<div class="gauge best" style="width: 14.48%;">Dar: 22.94 s</div>
<div class="gauge normal" style="width: 19.17%">Dar+sparse: 30.36 s</div>
<div class="gauge normal" style="width: 19.16%">Dar+sparse+binary delta: 30.35 s</div>
<div class="gauge normal" style="width: 99.63%">Rsync: 157.81 s</div>
<div class="gauge normal" style="width: 100%">Rsync+sparse: 158.39 s</div>
<div class="gauge normal" style="width: 16.87%">tar: 26.72 s </div>
<div class="gauge normal" style="width: 16.59%">tar+sparse: 26.27 s</div>
</div>
<h4>Execution time to restore a single differential backup</h4>
<div class="cadre">
<div class="gauge normal" style="width: 100%;">Dar: 3.48 s</div>
<div class="gauge normal" style="width: 100%">Dar+sparse: 3.48 s</div>
<div class="gauge normal" style="width: 98.85%">Dar+sparse+binary delta: 3.44 s</div>
<div class="gauge ref" style="width: 100%">Rsync: not supported</div>
<div class="gauge ref" style="width: 100%">Rsync+sparse: not supported</div>
<div class="gauge normal" style="width: 42.53%">tar: 1.48 s </div>
<div class="gauge normal" style="width: 43.1%">tar+sparse: 1.5 s</div>
</div>
<h4>Execution time to restore a whole system - full + differential backup</h4>
<p>
We use here the same extrapolation of a week of daily backup done above: the first backup being a full backup
and differential/incremental backups done the next days.
</p>
<p>
<u>Clarifying the terms used:</u> the <u>differential</u> backup saves only what has changed since the full backup
was made. The consequence is that each day the backup is slightlty bigger to process, depending on the
way data changed (if all files change every day, like mailboxes, user files, ...) each new differential backup will have the same size
and take the same processing time to complete.
At the opposite, if new data is added each day, the differential backup size will be each day the sum of the <i>incremental</i>
backups that could be done instead since the full backup was made.
</p>
<p>
At the difference of the differential backup, the <u>incremental</u> backup saves only what has changed
since the last backup (full or incremental). For constant activity like the steady Linux system we used here, the incremental
backup size should be the same along the time (and equivalent to the size of the first differential backup), thus the extrapolation is easy
and not questionable: the restoration time is the time to restore the full and the time to restore the first differential backup times
the number of days that passed.
</p>
<h4>Execution time to restore a whole system - lower bound</h4>
<p>
The lower bound, is the sum of the execution time of the restoration of the full backup and one differential backup
seen just above. It corresponds the minimum execution time restoring a whole system from full+differnential backup.
</p>
<div class="cadre">
<div class="gauge best" style="width: 16.68%;">Dar: 26.42 s</div>
<div class="gauge normal" style="width: 21.36%">Dar+sparse: 33.84 s</div>
<div class="gauge normal" style="width: 21.33%">Dar+sparse+binary delta: 33.79 s</div>
<div class="gauge ref" style="width: 99.63%">Rsync: full backup only 157.81 s</div>
<div class="gauge ref" style="width: 100%">Rsync+sparse: full backup only 158.39</div>
<div class="gauge normal" style="width: 17.80%">tar: 28.2 s</div>
<div class="gauge normal" style="width: 17.53%">tar+sparse: 27.77 s</div>
</div>
<h4>Execution time to restore a whole system - higher bound</h4>
<p>
The higher bound, is the sum of the execution time of the restoration plus seven times the execution time of the differential
backup. It corresponds the worse case scenario where each day new data is added (still using a steady Linux system with constant activity).
It also corresponds the scenario of restoring a whole system from a full+incremental backups (7 incremental backup have to be restored, in
that week span scenario):
</p>
<div class="cadre">
<div class="gauge normal" style="width: 29.86%;">Dar: 47.3 s</div>
<div class="gauge normal" style="width: 34.55%">Dar+sparse: 54.72 s</div>
<div class="gauge normal" style="width: 34.36%">Dar+sparse+binary delta: 54.43 s</div>
<div class="gauge ref" style="width: 99.63%">Rsync: full backup only 157.81 s</div>
<div class="gauge ref" style="width: 100%">Rsync+sparse: full backup only 158.39</div>
<div class="gauge normal" style="width: 23.41%">tar: 37.08 s </div>
<div class="gauge best" style="width: 23.21%">tar+sparse: 36.77 s</div>
</div>
<h4>Execution time to create a backup</h4>
<div class="cadre">
<div class="gauge normal" style="width: 81.62%;">Dar: 149.73 s</div>
<div class="gauge normal" style="width: 86.13%">Dar+sparse: 157.99 s</div>
<div class="gauge normal" style="width: 88.65%">Dar+sparse+binary delta: 162.62 s</div>
<div class="gauge normal" style="width: 85.58%">Rsync: 156.98 s</div>
<div class="gauge normal" style="width: 100%">Rsync+sparse: 183.44 s</div>
<div class="gauge best" style="width: 81%">tar: 148.59 s </div>
<div class="gauge normal" style="width: 81.43%">tar+sparse: 149.38 s</div>
</div>
<h4>Ciphering/deciphering performance</h4>
<p>
There is several reasons that implies the need of ciphering data:
</p>
<ul>
<li>if your disk is ciphered, would you store your backup in clear on the cloud?</li>
<li>do you trust your cloud provider to not inspect your data for marketing profiling?</li>
<li>Are you sure your patented data, secret industrial recipies will not be used by competition?</li>
<li>and so on</li>
</ul>
<p>
The ciphering execution time is independent on the nature of the backup, full or differential, compressed
or not. To evaluate the ciphering performance we will use the same data sets as previously, both compressed
and uncompressed. However not all software under test are able to cipher the resulting backup. <i>rsync</i>
is not able to do so.
</p>
<h5>Full backup+restoration execution time</h5>
<div class="cadre">
<div class="gauge normal" style="width: 100%">Dar: 9.13 s</div>
<div class="gauge ref" style="width: 100%">Rsync: N/A</div>
<div class="gauge best" style="width: 80.9%">Tar (openssl): 7.39 s</div>
</div>
<h5>Execution time for the restoration of a single file</h5>
<div class="cadre">
<div class="gauge best" style="width: 23.4%">Dar: 0.42 s</div>
<div class="gauge ref" style="width: 100%">Rsync: N/A</div>
<div class="gauge normal" style="width: 100%">Tar (openssl): 1.79 s</div>
</div>
<h5>Storage requirement ciphered without compression</h5>
<div class="cadre">
<div class="gauge best" style="width: 97.9%">Dar: 1.46 GiB</div>
<div class="gauge ref" style="width: 100%">Rsync: N/A</div>
<div class="gauge normal" style="width: 100%">Tar (openssl): 1.49 GiB</div>
</div>
<p>
See the corresponding <a href="benchmark_logs.html#backup_perf">test logs</a> for more details.
</p>
<h2>Conclusion</h2>
<p>
So far we have measured different perfomance aspects, evaluated available features, tested backup robusness and observed backup exhaustivity
of the different backup softwares under test. This gives a lot of
information already summarized above. But it would still not be of a great use to anyone reading this document
(espetially the one jumping to its conclusion ;^) ) so we have to get back to use cases and their respective requirements
to obtain the <a href="#oil">essential oil drop</a> anyone can use immediately:
</p>
<h3>Criteria for the different use cases</h3>
<div class=table>
<table>
<tr>
<th>Use Cases</th>
<th>Key Point</th>
<th>Optional interesting features</th>
</tr>
<tr>
<th>Local directory copy</th>
<td>
<ul>
<li>execution speed</li>
</ul>
</td>
<td>
<ul>
<li>completness of copied data and metadata</li>
</ul>
</td>
</tr>
<tr>
<th>remote directory copy - wide network</th>
<td>
<ul>
<li>execution speed</li>
</ul>
</td>
<td>
<ul>
<li>completness of copied data and metadata</li>
<li>on wire ciphering</li>
</ul>
</td>
</tr>
<tr>
<th>remote directory copy - narrow network</th>
<td>
<ul>
<li>execution speed</li>
<li>data reduction on wire</li>
</ul>
</td>
<td>
<ul>
<li>completness of copied data and metadata</li>
<li>on wire ciphering</li>
</ul>
</td>
</tr>
<tr>
<th>Full backups only</th>
<td>
<ul>
<li>completness of backed up data and metadata</li>
<li>data reduction on storage</li>
</ul>
</td>
<td>
<ul>
<li>fast restoration of a few files</li>
<li>fast restoration of a whole backup</li>
</ul>
</td>
</tr>
<tr>
<th>full+diff/incr. backup</th>
<td>
<ul>
<li>completness of backed up data and metadata</li>
<li>data reduced on storage</li>
</ul>
</td>
<td>
<ul>
<li>fast restoration of a few files</li>
<li>fast restoration of a whole backup</li>
<li>managing tool of backups rotation</li>
</ul>
</td>
</tr>
<tr>
<th>Archiving of private data</th>
<td>
<ul>
<li>data reduction on storage</li>
<li>robustness of the archive</li>
</ul>
</td>
<td>
<ul>
<li>ciphering</li>
<li>redundancy data</li>
</ul>
</td>
</tr>
<tr>
<th>Archiving of public data</th>
<td>
<ul>
<li>data reduction on storage</li>
<li>robustness of the archive</li>
</ul>
</td>
<td>
<ul>
<li>signing</li>
<li>fast decompression algorithm</li>
</ul>
</td>
</tr>
<tr>
<th>Private data exchange over Internet</th>
<td>
<ul>
<li>data reduction over the network</li>
<li>asymmetric encryption and signing</li>
</ul>
</td>
<td>
<ul>
<li>redundancy data</li>
<li>multi-volume backup/archive</li>
<li>integrated network protocols in backup tool</li>
</ul>
</td>
</tr>
<tr>
<th>Public data exchange over Internet</th>
<td>
<ul>
<li>data reduction over the network</li>
</ul>
</td>
<td>
<ul>
<li>hashing</li>
<li>sigining</li>
<li>integrated network protocols in backup tool</li>
</ul>
</td>
</tr>
</table>
</div>
<h3>Complementary criteria depending on the storage type</h3>
<p>
And depending on the target storage, the following adds on top:
</p>
<div class=table>
<table>
<tr>
<th>Use Cases</th>
<th>Key Point</th>
<th>Optional interesting features</th>
</tr>
<th>Local disk</th>
<td>
<ul>
<li>execution speed</li>
</ul>
</td>
<td>
<ul>
<li>hashing</li>
</ul>
</td>
<tr>
<th>Data stored on private NAS</th>
<td>
<ul>
<li>data reduction on storage</li>
</ul>
</td>
<td>
<ul>
<li>multi-volume backup</li>
<li>integrated network protocols in backup tool</li>
<li>ciphering</li>
</ul>
</td>
</tr>
<tr>
<th>Data stored on public cloud</th>
<td>
<ul>
<li>data reduction on storage and on wire</li>
<li>ciphering</li>
</ul>
</td>
<td>
<ul>
<li>multi-volumes backup</li>
<li>integrated network protocols in backup tool</li>
</ul>
</td>
</tr>
<tr>
<th>Data stored on removable media (incl. tapes)</th>
<td>
<ul>
<li>multi-volume backup</li>
<li>data reduction on storage</li>
<li>on-fly hashing</li>
</ul>
</td>
<td>
<ul>
<li>ciphering</li>
<li>redundancy data</li>
</ul>
</td>
</tr>
</table>
</div>
<h3>Essential <a name="oil">oil</a> drop</h3>
<p>
In summary, putting in front of these requirements the different measures we did:
</p>
<ul>
<li>exhasitivity of backed up data</li>
<li>available features around backup</li>
<li>backup robustness facing to media corruption</li>
<li>overall performance</li>
</ul>
<p>
We can summarize the best software to put in front of each particular use case:
</p>
<table class=center>
<tr>
<th>Use Cases</th>
<th>Local disk storage</th>
<th>Private NAS</th>
<th>Public Cloud</th>
<th>Removable media</th>
</tr>
<tr>
<th>Local directory copy</th>
<td>
<div class=optimum>cp</div>
<div class="limited tooltip">dar <span class=text>not the fastest</span></div>
<div class="limited tooltip">rsync <span class=text>not the fastest</span></div>
<div class="ideal tooltip">tar <span class=text>not the fastest</span></div>
</td>
<td>
-
</td>
<td>
-
</td>
<td>
-
</td>
</tr>
<tr>
<th>One time remote directory copy</th>
<td>
-
</td>
<td>
<div class=optimum>dar</div>
<div class="ideal tooltip">rsync<span class=text>not the fastest</span></div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="ideal tooltip">rsync<span class=text>not the fastest</span></div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="ideal tooltip">rsync<span class=text>not the fastest</span></div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
</tr>
<tr>
<th>Recurrent remote directory copy</th>
<td>
-
</td>
<td>
<div class="ideal tooltip">dar<span class=text>fastest but automation is a bit less straight forward than using <i>rsync</i></span></div>
<div class="optimum">rsync</div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>fastest but automation is a bit less straight forward than using <i>rsync</i></span></div>
<div class=optimum>rsync</div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>fastest but automation is a bit less straight forward than using <i>rsync</i></span></div>
<div class=optimum>rsync</div>
<div class="limited tooltip">tar<span class=text>no network protocol embedded</span></div>
</td>
</tr>
<tr>
<th>
Full backups only<br/>
(private data)
</th>
<td>
<div class="optimum tooltip">dar<span class=text>has the advantage to provide long historization of backups</span></div>
<div class="ideal tooltip">rsync<span class=text>no data reduction on storage, slow to restore a whole filesystem</span></div>
<div class="ideal tooltip">tar<span class=text>not saving all file attributes and inode types, slow to restore a few files</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>no data reduction on storage</span></div>
<div class="ideal tooltip">tar<span class=text>not saving all file attributes and inode types, slow to restore a few files, no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>no data ciphering and no reduction on storage</span></div>
<div class="limited tooltip">tar<span class=text>not embedded ciphering, not the strongest data encryption, not saving all file attributes and inode types, slow to restore a few files, no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>no multi-volume support, no data ciphering and no reduction on storage</span></div>
<div class="limited tooltip">tar<span class=text>compression and multi-volume are not supported at the same time, not saving all file attributes and inode types, not embedded ciphering, not the strongest data encryption</span></div>
</td>
</tr>
<tr>
<th>
full+diff/incr. backups<br/>
(priate data)
</th>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>differential backup not supported, full backup is overwritten</span></div>
<div class="ideal tooltip">tar<span class=text>not saving all file attributes and inode types, slow to restore a few files</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>differential backup not supported, full backup is overwritten</span></div>
<div class="ideal tooltip">tar<span class=text>not saving all file attributes and inode types, slow to restore a few files, no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>differential backup not supported, full backup is overwritten</span></div>
<div class="limited tooltip">tar<span class=text>not embedded ciphering, not the strongest data encryption, not saving all file attributes and inode types, slow to restore a few files, no network protocol embedded</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>differential backup not supported, full backup is overwritten, no support for multi-volime, no data reduction, no ciphering</span></div>
<div class="limited tooltip">tar<span class=text>compression and multi-volume are not supported at the same time, not saving all file attributes and inode types, not embedded ciphering, not the strongest data encryption</span></div>
</td>
</tr>
<tr>
<th>Archiving of private data</th>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>no data reduction on storage, no detection of data corruption, complex parity data addition</span></div>
<div class="ideal tooltip">tar<span class=text>no detection of data corruption or loss of all data after the first corruption met</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>no data reduction, no detection of data corruption, complex parity data addition</span></div>
<div class="ideal tooltip">tar<span class=text>no detection of data corruption or loss of all data after the first corruption met</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>no ciphering, no data reduction, no detection of data corruption, complex parity data addition</span></div>
<div class="limited tooltip">tar<span class=text>no detection of data corruption or loss of all data after the first corruption met, no embedded ciphering, no protection against plain-text attack</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>no data reduction, no multi-volume, no ciphering, no detection of data corruption, complex parity data addition</span></div>
<div class="noway tooltip">tar<span class=text>compression and multi-volume are not supported at the same time, no detection of data corruption or loss of all data after the first corruption met, no ciphering</span></div>
</td>
</tr>
<tr>
<th>Archiving of public data</th>
<td>
<div class="ideal tooltip">dar<span class=text>most robust format but not as standard as <i>tar</i>'s</span></div>
<div class="limited tooltip">rsync<span class=text>no reduction on storage</span></div>
<div class=optimum>tar</div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>most robust archive format but not as standard as <i>tar</i>'s</span></div>
<div class="noway tooltip">rsync<span class=text>no reduction on storage, complicated to download a directory tree and files from other protocols than rsync</span></div>
<div class=optimum>tar</div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>most robust archive format but not as standard as <i>tar</i></span></div>
<div class="noway tooltip">rsync<span class=text>no reduction on storage, complicated to download a directory tree and files from other protocols than rsync</span></div>
<div class=optimum>tar</div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>no reduction on storage, no multi-volume, no detection of data corruption, complex parity data addition</span></div>
<div class="limited tooltip">tar<span class=text>compression and multi-volume are not supported at the same time</span></div>
</td>
</tr>
<tr>
<th>Private data exchange over Internet</th>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>not the best data reduction over the network</span></div>
<div class="ideal tooltip">tar<span class=text>best data reduction on network but no embedded ciphering, no integrated network protocols</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="limited tooltip">rsync<span class=text>no data reduction on storage, not the best data reduction over the network</span></div>
<div class="ideal tooltip">tar<span class=text>best data reduction on network, but lack of embedded ciphering, lack of integrated network protocols</span></div>
</td>
<td>
<div class=optimum>dar</div>
<div class="noway tooltip">rsync<span class=text>no ciphering and no data reduction on storage</span></div>
<div class="limited tooltip">tar<span class=text>no embedded ciphering, no integrated network protocols, no protection against plain-text attack, only old KDF functions supported, complex and error prone use of openssl to cipher the archive</span></div>
</td>
<td>
-
</td>
</tr>
<tr>
<th>Public data exchange over Internet</th>
<td>
<div class="ideal tooltip">dar<span class=text>not the best data reduction over the network</span></div>
<div class="ideal tooltip">rsync<span class=text>not the best data reduction over the network</span></div>
<div class=optimum>tar</div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>not the best data reduction over the network</span></div>
<div class="limited tooltip">rsync<span class=text>no data reduction on storage, not the best data reduction over the network</span></div>
<div class=optimum>tar</div>
</td>
<td>
<div class="ideal tooltip">dar<span class=text>not the best data reduction over the network</span></div>
<div class="limited tooltip">rsync<span class=text>no data reduction on storage, not the best data reduction over the network</span></div>
<div class=optimum>tar</div>
</td>
<td>
-
</td>
</tr>
</table>
<p>
In each cell of the previous table, the different softwares are listed in alphabetical order, they get colorized according to the following code:
</p>
<div class=table>
<table class=center>
<tr>
<th style="width: 20%; padding: 10px">Color codes</th>
<td style="width: 10%; padding: 10px"><div class=optimum>best solution</div></td>
<td style="width: 10%; padding: 10px"><div class=ideal>good solution</div></td>
<td style="width: 10%; padding: 10px"><div class=limited>not optimal</div></td>
<td style="width: 10%; padding: 10px"><div class=noway>not adapted</div></td>
</tr>
</table>
</div>
<p>
Hovering the mouse on a particular item gives more details about the reason it has not been selected as the best solution for a particular need.
</p>
</div>
</body>
</html>
|
