Can you try the following command:
$ time dd if=/dev/i500-lto4-tape1 bs=32k of=/dev/i500-lto4-tape2
Put an amanda tape in /dev/i500-lto4-tape1 and a new tape in
/dev/i500-lto4-tape2
Run the command a few times.
You can also try to use bigger block size 128k-256k
Jean-Louis
On 03/14/2012 09:04 AM, Sven Rudolph wrote:
Jean-Louis Martineau<[email protected]> writes:
Try to set a larger value for 'device-output-buffer-size'.
Tried, didn't really help:
device-output-buffer-size 128m
(I append the nearly-complete amanda.conf below.)
Amanda now continues with the following tape, beacuse I set partsize:
define tapetype TERTIARY {
LTO4
#part-size 0
part-size 5g
}
(Originally I wanted a mode where it doesnt split in parts but only
splits on LEOM. Worked with amanda 3.2.x and erlier, I didn't get it
running on 3.3.x)
This change also shows that sometimes parts are written fast:
Wed Mar 14 13:22:00 2012: amvault: Building type SPLIT_FILE header of
32768-32768 bytes with name='fileserver' disk='/data' dumplevel=0 and
blocksize=32768
Wed Mar 14 13:22:49 2012: amvault: Building type SPLIT_FILE header of
32768-32768 bytes with name='fileserver' disk='/data' dumplevel=0 and
blocksize=32768
Wed Mar 14 13:38:49 2012: amvault: Building type SPLIT_FILE header of
32768-32768 bytes with name='fileserver' disk='/data' dumplevel=0 and
blocksize=32768
Wed Mar 14 13:50:45 2012: amvault: Building type SPLIT_FILE header of
32768-32768 bytes with name='fileserver' disk='/data' dumplevel=0 and
blocksize=32768
So it works now, but still very slow. Will take more than 2 weeks. And
probably does brilliant shoeshining.
Sven
On 02/27/2012 11:07 AM, Sven Rudolph wrote:
(I sent this message on Feb 23, but it didn't appear on the list. I
resend it here without the logfiles attached. Maybe they were too
large.)
Hello,
I am trying to setup amvault for tape-to-tape duplication.
My problem is that the write speed is very low (about 12 MB/s) whereas
the native write speed usually is about 115 MB/s:
PART taper Set1K-2011-01 2 fileserver /backup 20120101030850 1/-1 0 [sec
16429.995803 bytes 208375369542 kps 12385.369628]
DONE taper fileserver /backup 20120101030850 1 0 [sec 16430.000000 bytes
208375369542 kps 12385.366464 orig-kb 310011210]
The amreport mail message finally says:
Avg Tp Write Rate (k/s) 14148.8 14148.8 --
USAGE BY TAPE:
Label Time Size % DLEs Parts
Set1K-2011-01 14:36 726591M 94.4 3 3
The tape drives ar IBM FC LTO-4, each connected to its own 4Gb FC
port. Running dd's to each drive in parallel I do not see a
bottleneck:
$ dd if=/dev/i500-lto4-tape1 bs=32k of=/dev/null bs=32k count=1000000
1000000+0 records in
1000000+0 records out
32768000000 bytes (33 GB) copied, 281.63 s, 116 MB/s
$ dd if=/dev/i500-lto4-tape2 bs=32k of=/dev/null bs=32k count=1000000
1000000+0 records in
1000000+0 records out
32768000000 bytes (33 GB) copied, 310.629 s, 105 MB/s
I tried to strace the processes/threads:
# ls -l /proc/21570/fd
total 0
lrwx------ 1 backup backup 64 Feb 23 14:07 0 -> /dev/pts/6
lrwx------ 1 backup backup 64 Feb 23 14:07 1 -> /dev/pts/6
l-wx------ 1 backup backup 64 Feb 23 14:07 10 ->
/var/log/amanda/server/MonthlySet1/amvault.20120223133643.debug
lrwx------ 1 backup backup 64 Feb 23 13:38 2 -> /dev/pts/6
lrwx------ 1 backup backup 64 Feb 23 14:07 3 -> anon_inode:[eventfd]
lrwx------ 1 backup backup 64 Feb 23 14:07 5 -> /dev/nst3
lr-x------ 1 backup backup 64 Feb 23 14:07 6 -> pipe:[5827058]
l-wx------ 1 backup backup 64 Feb 23 14:07 7 -> pipe:[5827058]
lrwx------ 1 backup backup 64 Feb 23 14:07 8 -> /dev/nst2
# ls -l /proc/21570/task/
total 0
dr-xr-xr-x 5 backup backup 0 Feb 23 13:45 21570
dr-xr-xr-x 5 backup backup 0 Feb 23 13:45 21573
dr-xr-xr-x 5 backup backup 0 Feb 23 13:45 21800
dr-xr-xr-x 5 backup backup 0 Feb 23 13:45 21801
My summary of whats going on (full logs attached below):
14:18:17.778803 the reader process starts reading 32k blocks
14:18:18.018007 the writer process starts writing
14:18:18.017894 the reader process doesnt show all the futex calls
14:18:18.285803 the writer stopps writing
14:18:18.295847 the reader stops reading
14:18:24.281473 the writer starts writing
14:18:24.281608 the reader starts reading
...
So ... something seems to be to slow so that the tape drives don't
keep streaming.
I am using Amanda 3.3.0 on Debian.
Any ideas, suggestions, ... ?
-----
Here my amanda.conf:
org "MonthlySet1" # your organization name for reports
mailto "ambackup" # space separated list of operators at your site
dumpuser "backup" # the user to run dumps under
inparallel 15 # maximum dumpers that will run in parallel
netusage 130000 # maximum net bandwidth for Amanda, in KB per sec
maxdumps 1 # amount of parallelism per client
dumporder "BSSTSsSsSS"
etimeout 3600 # time per disk on a client planner waits for estimates
dtimeout 20000 # time the dumper waits for input data
dumpcycle 4 weeks # the number of days in the normal dump cycle
runspercycle 5
tapecycle 64 tapes # the number of tapes in rotation
bumpsize 20 MB # minimum savings (threshold) to bump level 1 -> 2
bumpdays 1 # minimum days at each level
bumpmult 4 # threshold = bumpsize * (level-1)**bumpmult
runtapes 18 # explained in WHATS.NEW
device-output-buffer-size 128m
displayunit "m"
define changer i500-changer {
tapedev "chg-robot:/dev/i500-changer"
property "tape-device" "0=tape:/dev/i500-lto4-tape"
property "use-slots" "1-22"
device_property "LEOM" "yes"
}
tapedev "i500-changer"
define changer i500-changer-vault {
tapedev "chg-robot:/dev/i500-changer"
property "tape-device" "1=tape:/dev/i500-lto4-tape2"
property "use-slots" "87-108"
device_property "LEOM" "yes"
}
amrecover_changer "changer"
autoflush off
#taperalgo largestfit
taperalgo largest
flush-threshold-dumped 300
taperflush 0
# # flush-threshold-scheduled>= flush-threshold-dumped
# # flush-threshold-scheduled>= taperflush
flush-threshold-scheduled 500
tapetype LTO4
labelstr
"^((Set1-(Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov|20..)-)|(Set1K-20..-)).*$"
# label constraint regex: all tapes must match
columnspec
"HostName=0:8,Disk=1:18,DumpRate=1:6,OrigKB=1:8,OutKB=1:8,TapeRate=1:6"
reserve 1 # holding-disk space reserved for non-full backups
holdingdisk holdingdisk1 {
directory "/var/spool/amanda/1/1m"
use -1GB
chunksize 1 GB
}
# ... more holdingdisks
infofile "/var/lib/amanda/MonthlySet1/curinfo" # database filename
logdir "/var/log/amanda/MonthlySet1/" # log filename
indexdir "/var/lib/amanda/MonthlySet1/index"
define tapetype LTO4 {
comment "LTO-4"
length 770000 mbytes
# length 790000 mbytes
filemark 0 kbytes
speed 115000 kps
lbl-templ "/etc/amanda/WeeklySet1/LTO-label.ps"
}
define tapetype TERTIARY {
LTO4
#part-size 0
part-size 5g
}
# ... dumptypes ...
