./backups/rsync-vs-rdiff-backup.txt
download original
Comparing rsnapshot/rsync and rdiff-backup, using a ~14G Virtualbox
VMs directory (so we have some big binaries that change a little every
time -- a scenario rdiff-backup should be better at) as well as a
larger number of smallish files.
Source (/home/olaf/VirtualBox VMs/) is an internal SATA-2 hard drive,
destination (/mnt/backup1/test) is a USB-2 connected external hard
drive with a dm-crypt/LUKS1 encrypted ext4 filesystem. iotop and some
simple tests indicate about 65 MB/s read bandwidth for the source and
30 MB/s write bandwidth for the destination.
## start with the source containing only the Virtualbox data, not the
many small files
# prime the buffer cache
olaf@tack:~$ find VirtualBox\ VMs/ -type f -print0 | xargs -0 cat >/dev/null
# initial backup -- rsync
tack:/mnt/backup1/test# time rsync -a /home/olaf/VirtualBox\ VMs/ rsync/bup.1
real 7m50.050s
user 1m21.448s
sys 0m59.796s
tack:/mnt/backup1/test#
# initial backup -- rdiff-backup
tack:/mnt/backup1/test# time rdiff-backup /home/olaf/VirtualBox\ VMs/ rdiffbackup/bup
real 7m46.949s
user 0m35.204s
sys 0m29.980s
tack:/mnt/backup1/test#
olaf@tack:~$ virtualbox
(launch WinXP VM, create file foo.txt on Desktop, open in notepad,
write "Hello World" into it, save, leave notepad running, save&exit VM)
olaf@tack:~$
olaf@tack:~$
olaf@tack:~$ find VirtualBox\ VMs/ -cmin -2 -o -mmin -2
VirtualBox VMs/NewXP
VirtualBox VMs/NewXP/Logs
VirtualBox VMs/NewXP/Logs/VBox.log.3
VirtualBox VMs/NewXP/Logs/VBox.log.1
VirtualBox VMs/NewXP/Logs/VBox.log
VirtualBox VMs/NewXP/Logs/VBox.log.2
VirtualBox VMs/NewXP/NewXP.vbox-prev
VirtualBox VMs/NewXP/NewXP.vbox
VirtualBox VMs/NewXP/Snapshots
VirtualBox VMs/NewXP/Snapshots/2014-08-10T02-30-17-707740000Z.sav
VirtualBox VMs/NewXP/Snapshots/{b8630efa-1eef-48f4-872d-28fa6f86167f}.vdi
olaf@tack:~$
# 1st incremental backup -- rsync
tack:/mnt/backup1/test# time rsync -a --delete --link-dest=../bup.1 /home/olaf/VirtualBox\ VMs/ rsync/bup.2
real 1m51.257s
user 0m20.528s
sys 0m15.020s
tack:/mnt/backup1/test#
# 1st incremental backup -- rdiff-backup
tack:/mnt/backup1/test# time rdiff-backup /home/olaf/VirtualBox\ VMs/ rdiffbackup/bup
real 9m0.157s
user 1m2.692s
sys 0m15.500s
tack:/mnt/backup1/test#
# size comparison:
tack:/mnt/backup1/test# du -hs rdiffbackup
13G rdiffbackup
tack:/mnt/backup1/test# du -hs rsync/
16G rsync/
## ok, apparently rdiff-backup did indeed create binary diffs and save
space, but took over 4 times as long as rsync to complete
# for the 2nd incremental backup, add two directories with a bunch of
mostly smallish files -- my "utils" and "mydocs" repositories
## copy it in, clean, checkout revision from January 1st, 2012
olaf@tack:~$ cp -rf utils VirtualBox\ VMs/
olaf@tack:~$ cd VirtualBox\ VMs/utils/
olaf@tack:~/VirtualBox VMs/utils$ git ls-files -o -z | xargs -0 rm -f
olaf@tack:~/VirtualBox VMs/utils$ git checkout -- .
olaf@tack:~/VirtualBox VMs/utils$ git checkout `git rev-list -n 1 --before="2012-01-01" master`
Note: checking out '1b3d958ebebffd44641c66b606796d02e8415339'.
You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by performing another checkout.
If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -b with the checkout command again. Example:
git checkout -b new_branch_name
HEAD is now at 1b3d958... myhomehost/setup-6to4: bugfix (set up correct route into local /64 network)
olaf@tack:~/VirtualBox VMs/utils$
olaf@tack:~/VirtualBox VMs$ git clone ~/doc/mydocs
Cloning into 'mydocs'...
done.
Checking out files: 100% (2037/2037), done.
olaf@tack:~/VirtualBox VMs$ cd mydocs/
olaf@tack:~/VirtualBox VMs/mydocs$ git checkout `git rev-list -n 1 --before="2012-01-01" master`
Note: checking out '1d37aa4906dcf915dd64d0b39591b5bc3db13c50'.
You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by performing another checkout.
If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -b with the checkout command again. Example:
git checkout -b new_branch_name
HEAD is now at 1d37aa4... yt-posts merged
olaf@tack:~/VirtualBox VMs/mydocs$
olaf@tack:~/VirtualBox VMs/mydocs$ cd ..
olaf@tack:~/VirtualBox VMs$
olaf@tack:~/VirtualBox VMs$
olaf@tack:~/VirtualBox VMs$
olaf@tack:~/VirtualBox VMs$ du -hs utils/ mydocs/
2.6M utils/
114M mydocs/
olaf@tack:~/VirtualBox VMs$
olaf@tack:~/VirtualBox VMs$ find utils/ -type f | wc -l
405
olaf@tack:~/VirtualBox VMs$ find mydocs/ -type f | wc -l
2304
olaf@tack:~/VirtualBox VMs$
# 2nd incremental backup -- rsync
tack:/mnt/backup1/test# time rsync -a --delete --link-dest=../bup.2 /home/olaf/VirtualBox\ VMs/ rsync/bup.3
real 0m5.669s
user 0m0.696s
sys 0m0.704s
tack:/mnt/backup1/test#
# 2nd incremental backup -- rdiff-backup
tack:/mnt/backup1/test# time rdiff-backup /home/olaf/VirtualBox\ VMs/ rdiffbackup/bup
real 1m42.415s
user 0m3.076s
sys 0m1.212s
tack:/mnt/backup1/test#
# for the 3rd incremental backup, forward the versions of the two
directories checked out in the previous step by two years
olaf@tack:~/VirtualBox VMs$ cd utils/
olaf@tack:~/VirtualBox VMs/utils$
olaf@tack:~/VirtualBox VMs/utils$
olaf@tack:~/VirtualBox VMs/utils$ git checkout `git rev-list -n 1 --before="2014-01-01" master`
Previous HEAD position was 1b3d958... myhomehost/setup-6to4: bugfix (set up correct route into local /64 network)
HEAD is now at 163f5e9... myhomehost/reinit-network
olaf@tack:~/VirtualBox VMs/utils$ cd ..
olaf@tack:~/VirtualBox VMs$ cd mydocs/
olaf@tack:~/VirtualBox VMs/mydocs$ git checkout `git rev-list -n 1 --before="2014-01-01" master`
Previous HEAD position was 1d37aa4... yt-posts merged
HEAD is now at 28da1ff... passwords update
olaf@tack:~/VirtualBox VMs/mydocs$
# 3rd incremental backup -- rsync
tack:/mnt/backup1/test# time rsync -a --delete --link-dest=../bup.3 /home/olaf/VirtualBox\ VMs/ rsync/bup.4
real 0m0.415s
user 0m0.088s
sys 0m0.164s
tack:/mnt/backup1/test#
# 3rd incremental backup -- rdiff-backup
tack:/mnt/backup1/test# time rdiff-backup /home/olaf/VirtualBox\ VMs/ rdiffbackup/bup
real 0m14.533s
user 0m1.636s
sys 0m0.240s
tack:/mnt/backup1/test#
# quick checks for correctness of the backups (latest snapshot)
tack:/mnt/backup1/test# diff -ur rdiffbackup/bup/utils/ /home/olaf/VirtualBox\ VMs/utils/
tack:/mnt/backup1/test# diff -ur rsync/bup.4/utils/ /home/olaf/VirtualBox\ VMs/utils/
tack:/mnt/backup1/test#
tack:/mnt/backup1/test# diff -ur rdiffbackup/bup/mydocs/ /home/olaf/VirtualBox\ VMs/mydocs/
tack:/mnt/backup1/test# diff -ur rsync/bup.4/mydocs/ /home/olaf/VirtualBox\ VMs/mydocs/
tack:/mnt/backup1/test#
tack:/mnt/backup1/test# diff rsync/bup.NewXP/Snapshots/\{b8630efa-1eef-48f4-872d-28fa6f86167f\}.vdi /home/olaf/VirtualBox\ VMs/NewXP/Snapshots/\{b8630efa-1eef-48f4-872d-28fa6f86167f\}.vdi
bup.1/ bup.2/ bup.3/ bup.4/
tack:/mnt/backup1/test# diff rsync/bup.4/NewXP/Snapshots/\{b8630efa-1eef-48f4-872d-28fa6f86167f\}.vdi /home/olaf/VirtualBox\ VMs/NewXP/Snapshots/\{b8630efa-1eef-48f4-872d-28fa6f86167f\}.vdi
tack:/mnt/backup1/test#
## 4th incremental backup preparation -- start XP in virtualbox again,
start Chrome in XP, surf a couple of news sites, watch 2 short YT
videos, download
http://ftp.de.debian.org/debian/pool/main/e/emacs23/emacs23_23.4+1-4_amd64.deb,
leave browser running, save&exit VM
olaf@tack:~$ find VirtualBox\ VMs/ -type f -a \( -cmin -30 -o -mmin -30 \) -print0 | xargs -0 ls -l
-rw------- 1 olaf olaf 87151 Aug 10 06:05 VirtualBox VMs/NewXP/Logs/VBox.log
-rw------- 1 olaf olaf 85350 Aug 10 04:30 VirtualBox VMs/NewXP/Logs/VBox.log.1
-rw------- 1 olaf olaf 85038 Aug 10 03:50 VirtualBox VMs/NewXP/Logs/VBox.log.2
-rw------- 1 olaf olaf 85796 Oct 22 2013 VirtualBox VMs/NewXP/Logs/VBox.log.3
-rw------- 1 olaf olaf 88167 Aug 10 06:05 VirtualBox VMs/NewXP/NewXP.vbox
-rw------- 1 olaf olaf 88430 Aug 10 06:05 VirtualBox VMs/NewXP/NewXP.vbox-prev
-rw------- 1 olaf olaf 330667948 Aug 10 06:05 VirtualBox VMs/NewXP/Snapshots/2014-08-10T04-05-25-278196000Z.sav
-rw------- 1 olaf olaf 3303059456 Aug 10 06:05 VirtualBox VMs/NewXP/Snapshots/{b8630efa-1eef-48f4-872d-28fa6f86167f}.vdi
olaf@tack:~$
# 4th incremental backup -- rsync
tack:/mnt/backup1/test# time rsync -a --delete --link-dest=../bup.4 /home/olaf/VirtualBox\ VMs/ rsync/bup.5
real 1m57.103s
user 0m21.028s
sys 0m16.620s
tack:/mnt/backup1/test#
# 4th incremental backup -- rdiff-backup
tack:/mnt/backup1/test# time rdiff-backup /home/olaf/VirtualBox\ VMs/ rdiffbackup/bup
real 9m26.591s
user 1m35.844s
sys 0m16.296s
tack:/mnt/backup1/test#
size comparison:
tack:/mnt/backup1/test# du -hs rdiffbackup/
14G rdiffbackup/
tack:/mnt/backup1/test#
tack:/mnt/backup1/test# du -hs rsync/
20G rsync/
tack:/mnt/backup1/test#
# try to estimate the size of the binary deltas (I didn't do du
without -h after earlier backups)
tack:/mnt/backup1/test# ls -lh rdiffbackup/bup/rdiff-backup-data/increments/NewXP/Snapshots/
total 227M
-rw------- 1 olaf olaf 92M Aug 10 03:50 2014-08-10T01-50-38-582549000Z.sav.2014-08-10T04:18:12+02:00.snapshot.gz
-rw------- 1 root root 0 Aug 10 04:46 2014-08-10T02-30-17-707740000Z.sav.2014-08-10T04:18:12+02:00.missing
-rw------- 1 olaf olaf 93M Aug 10 04:30 2014-08-10T02-30-17-707740000Z.sav.2014-08-10T05:26:50+02:00.snapshot.gz
-rw------- 1 root root 0 Aug 10 06:12 2014-08-10T04-05-25-278196000Z.sav.2014-08-10T05:26:50+02:00.missing
-rw------- 1 olaf olaf 1.7M Aug 10 03:50 {b8630efa-1eef-48f4-872d-28fa6f86167f}.vdi.2014-08-10T04:18:12+02:00.diff.gz
-rw------- 1 olaf olaf 42M Aug 10 04:30 {b8630efa-1eef-48f4-872d-28fa6f86167f}.vdi.2014-08-10T05:26:50+02:00.diff.gz
tack:/mnt/backup1/test#
(the Aug 10 04:30 things should be the 4th backup)
==> summary
The rdiff-backup compressed binary delta thing works and saves a lot
of storage space on incremental backups (at least 90..95% for
Virtualvox VM images with few changes in them).
rsync/rsnapshot operated ~5..20 times faster than rdiff-backup for
incremental backups (for full/initial backups, running times are
almost the same):
For full backups, the effective transfer rate for both tools was ~30
MB/s -- close to to physical bandwith of the involved storage media.
For incremental backups:
When the backup contained about 115MB of new, smallish files in a
previously backed-up ~13GB directory, the effective transfer rate was
about 20 MB/s for rsync and 1.1 MB/s for rdiff-backup (see 2nd
incremental backup above).
When the backup contained about 3.6GB of modified large binaries in a
previously backed-up ~14GB directory (mostly just two files -- 300MB
and 3.3GB, with a binary delta size of about 130MB according to
rdiff-backup), the effective transfer rate, in relation to the size of
the modified files, was ~30 MB/s for rsync and 6.3 MB/s for
rdiff-backup -- or 1.1 MB/s (rsync) and 230 kB/s (rdiff-backup) in
relation to the size of the delta. (see 4th incremental backup above).
back to backups
(C) 1998-2017 Olaf Klischat <olaf.klischat@gmail.com>