libgocryptfs/Documentation/performance.txt

All test performed on a Pentium G630 (Dual-core Sandy Bridge, no AES-NI).
The working directory is on tmpfs.
The untar test uses https://cdn.kernel.org/pub/linux/kernel/v3.0/linux-3.0.tar.gz .
The archive is placed on tmpfs as well.

WRITE: dd if=/dev/zero of=zero bs=131072 count=2000
READ:  dd if=zero of=/dev/null bs=131072 count=2000 2>&1
UNTAR: time tar xzf ../linux-3.0.tar.gz
MD5:   time md5sum --quiet -c linux-3.0.md5sums
LS:    time ls -lR linux-3.0 > /dev/null
RM:    time rm -Rf linux-3.0

(or just run benchmark.bash)

VERSION           WRITE    READ     UNTAR  MD5   LS     RM    ENV CHANGE?
v0.4                                48          1.5    5
v0.5-rc1                            56          7     19
v0.5-rc1-1                          54          4.1    9
v0.5-rc1-2                          45          1.7    3.4
v0.6                                47          1.8    4.3
v0.7                                43          1.7    4.3
v0.7.2                              26          1.8    4.3
v0.8                                23          1.8    4.3
v0.9-rc2           94               24          1.8    4.5
v0.9               94               24          1.8    4.5
v0.11             104               22          1.7    4.5
v1.1              104               20          1.5    3.4    go1.7.1
v1.1.1-34         112               22          1.5    3.6    go1.7.3
v1.2.1-33         112               21    12    1.6    4.4    go1.8
-serialize_reads  116               21    39    1.5    4.4    (v1.2.1-33 with -serialize_reads)
v1.3-27           113               20    11    1.4    4.2
v1.3-53-gf44902a  119               19    12    1.6    4.1
v1.3-64-g80516ed  123               19    11    1.3    4.2
v1.3-67-g9837cb0  125               19    11    1.4    4.2    go1.8.3, Linux 4.10
v1.3-69-ge52594d  145               19.0  11.6  1.4    4.1
v1.4-1-g3c6fe98   154               17.2  11.7  1.4    4.1
v1.4-5-g0cc6f53   182      144      16.7  11.1  1.3    3.3
v1.4-8-g80676c6   178      148      16.1  11.0  1.3    4.0
v1.4-14-g9f4bd76  182      286      15.4   7.5  1.3    4.1
v1.4-45-gd5671b7  183      282      14.9   7.3  1.1    2.9
v1.4-45-gd5671b7  252      285      15.5   7.2  1.1    2.9    go1.8.3, Linux 4.11
v1.4.1            253      285      16.0   7.4  1.3    3.0    go1.9,   Linux 4.12.5
v1.4.1-6-g276567e 258      289      16.1   7.5  1.3    3.0
v1.5              228      292      17.6   9.3  1.5    3.5    go1.10.2, Linux 4.16.8
v1.6              250      289      17.7   8.0  1.3    3.2    go1.10.3, Linux 4.17.12

Results for EncFS for comparison (benchmark.bash -encfs):

VERSION            WRITE    UNTAR   MD5    LS     RM
encfs v1.9.1       95          20     8    2.8    3.8
Update performance.txt with new numbers 2017-06-29 19:00:16 +02:00			`All test performed on a Pentium G630 (Dual-core Sandy Bridge, no AES-NI).`
			`The working directory is on tmpfs.`
benchmark.bash: double write length Writing 1000 128KB blocks takes only 1 second and yielded inconsistent results. With 2000, things look saner. 2016-11-26 12:36:55 +01:00			`The untar test uses https://cdn.kernel.org/pub/linux/kernel/v3.0/linux-3.0.tar.gz .`
Update performance data for v0.7 2015-12-20 15:39:40 +01:00			`The archive is placed on tmpfs as well.`
OpenDir performance: Read DirIV once and reuse it for all names Formerly, we called decryptPath for every name. That resulted in a directory walk that reads in all diriv files on the way. Massive improvement for RM and LS (check performance.txt for details) VERSION UNTAR RM LS v0.4 48 5 1.5 v0.5-rc1 56 19 7 v0.5-rc1-1 54 9 4.1 <---- THIS VERSION 2015-11-29 19:57:48 +01:00
performance.txt: specify READ tests 2017-09-10 18:42:06 +02:00			`WRITE: dd if=/dev/zero of=zero bs=131072 count=2000`
			`READ: dd if=zero of=/dev/null bs=131072 count=2000 2>&1`
benchmark.bash: double write length Writing 1000 128KB blocks takes only 1 second and yielded inconsistent results. With 2000, things look saner. 2016-11-26 12:36:55 +01:00			`UNTAR: time tar xzf ../linux-3.0.tar.gz`
benchmark: add md5sum read performance benchmark 2017-03-18 16:23:33 +01:00			`MD5: time md5sum --quiet -c linux-3.0.md5sums`
Update performance data for v0.7 2015-12-20 15:39:40 +01:00			`LS: time ls -lR linux-3.0 > /dev/null`
			`RM: time rm -Rf linux-3.0`
OpenDir performance: Read DirIV once and reuse it for all names Formerly, we called decryptPath for every name. That resulted in a directory walk that reads in all diriv files on the way. Massive improvement for RM and LS (check performance.txt for details) VERSION UNTAR RM LS v0.4 48 5 1.5 v0.5-rc1 56 19 7 v0.5-rc1-1 54 9 4.1 <---- THIS VERSION 2015-11-29 19:57:48 +01:00
performance.txt: add numbers for current gocryptfs master and encfs v1.9.1 2016-11-26 12:46:58 +01:00			`(or just run benchmark.bash)`
Update performance.txt for v1.1 release 2016-10-09 23:19:08 +02:00
docs: label "ENV CHANGE" column 2017-08-15 19:03:07 +02:00			`VERSION WRITE READ UNTAR MD5 LS RM ENV CHANGE?`
Update performance.txt with new numbers 2017-06-29 19:00:16 +02:00			`v0.4 48 1.5 5`
			`v0.5-rc1 56 7 19`
			`v0.5-rc1-1 54 4.1 9`
			`v0.5-rc1-2 45 1.7 3.4`
			`v0.6 47 1.8 4.3`
			`v0.7 43 1.7 4.3`
			`v0.7.2 26 1.8 4.3`
			`v0.8 23 1.8 4.3`
			`v0.9-rc2 94 24 1.8 4.5`
			`v0.9 94 24 1.8 4.5`
			`v0.11 104 22 1.7 4.5`
docs: label "ENV CHANGE" column 2017-08-15 19:03:07 +02:00			`v1.1 104 20 1.5 3.4 go1.7.1`
			`v1.1.1-34 112 22 1.5 3.6 go1.7.3`
			`v1.2.1-33 112 21 12 1.6 4.4 go1.8`
			`-serialize_reads 116 21 39 1.5 4.4 (v1.2.1-33 with -serialize_reads)`
Update performance.txt with new numbers 2017-06-29 19:00:16 +02:00			`v1.3-27 113 20 11 1.4 4.2`
			`v1.3-53-gf44902a 119 19 12 1.6 4.1`
			`v1.3-64-g80516ed 123 19 11 1.3 4.2`
docs: label "ENV CHANGE" column 2017-08-15 19:03:07 +02:00			`v1.3-67-g9837cb0 125 19 11 1.4 4.2 go1.8.3, Linux 4.10`
Update performance.txt with new numbers 2017-06-29 19:00:16 +02:00			`v1.3-69-ge52594d 145 19.0 11.6 1.4 4.1`
			`v1.4-1-g3c6fe98 154 17.2 11.7 1.4 4.1`
			`v1.4-5-g0cc6f53 182 144 16.7 11.1 1.3 3.3`
Update performance.txt with new numbers 2017-06-29 23:45:39 +02:00			`v1.4-8-g80676c6 178 148 16.1 11.0 1.3 4.0`
Update performance.txt with new results Massive speed boost for streaming reads. 2017-07-01 10:00:50 +02:00			`v1.4-14-g9f4bd76 182 286 15.4 7.5 1.3 4.1`
Update performance.txt for to Getdents change 2017-08-15 19:07:08 +02:00			`v1.4-45-gd5671b7 183 282 14.9 7.3 1.1 2.9`
Update performance.txt for Linux kernel upgrade For some reason, writing became a lot faster in Linux 4.11 (scheduler improvements?). 2017-08-15 19:13:00 +02:00			`v1.4-45-gd5671b7 252 285 15.5 7.2 1.1 2.9 go1.8.3, Linux 4.11`
performance.txt: update for v1.4.1 latest commits, and kernel update $ uname -a Linux brikett 4.12.5-300.fc26.x86_64 #1 SMP Mon Aug 7 15:27:25 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux 2017-09-03 14:25:01 +02:00			`v1.4.1 253 285 16.0 7.4 1.3 3.0 go1.9, Linux 4.12.5`
			`v1.4.1-6-g276567e 258 289 16.1 7.5 1.3 3.0`
Update README and MANPAGE for v1.5 Also update the performance numbers. I see some slowdown, reason is not yet clear, but nothing to block the release. 2018-06-10 19:30:10 +02:00			`v1.5 228 292 17.6 9.3 1.5 3.5 go1.10.2, Linux 4.16.8`
Update README for v1.6 release 2018-08-18 12:19:47 +02:00			`v1.6 250 289 17.7 8.0 1.3 3.2 go1.10.3, Linux 4.17.12`
performance.txt: add numbers for current gocryptfs master and encfs v1.9.1 2016-11-26 12:46:58 +01:00
performance.txt: update for v1.4.1 latest commits, and kernel update $ uname -a Linux brikett 4.12.5-300.fc26.x86_64 #1 SMP Mon Aug 7 15:27:25 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux 2017-09-03 14:25:01 +02:00			`Results for EncFS for comparison (benchmark.bash -encfs):`
performance.txt: add numbers for current gocryptfs master and encfs v1.9.1 2016-11-26 12:46:58 +01:00
benchmark: add md5sum read performance benchmark 2017-03-18 16:23:33 +01:00			`VERSION WRITE UNTAR MD5 LS RM`
			`encfs v1.9.1 95 20 8 2.8 3.8`