Hi everyone-
I have an DLT-S4 drive hooked up to a Dell 1950 running CentOS and
Bacula 2.0.3. Everything works just fine, but the throughput to the tape
drive is really slow. After mucking around with tar, I found that
increasing the block size really improves the performance of the tape
drive. According to tapeinfo, the drive can handle block sizes to just
under 16 MB. Using a block size close to 16 MB gives me some of that
awesome speed that Quantum promised:
bash-3.00# tar -c --totals -b 32000 -f /dev/nst0 .
Total bytes written: 1179648000 (1.1GiB, 94MiB/s)
I went to change the Maximum block size value in my SD, but I get data
overrun errors while running a job with anything larger than 512 KB.
Here's some output from the SD when the block size is set to 900000:
bash-3.00# /usr/sbin/bacula-sd -f -d500
.. snip ...
cfile-prod-bkp-1-sd: append.c:300 back from write_end_session_label()
cfile-prod-bkp-1-sd: spool.c:137 Committing spooled data
cfile-prod-bkp-1-sd: spool.c:220 Despooling data
cfile-prod-bkp-1-sd: device.c:345 lock 0 from spool.c:235
cfile-prod-bkp-1-sd: block.c:443 binbuf=899991 buf_len=900000
cfile-prod-bkp-1-sd: block.c:443 binbuf=899993 buf_len=900000
cfile-prod-bkp-1-sd: block.c:443 binbuf=899995 buf_len=900000
cfile-prod-bkp-1-sd: dev.c:1655 weof_dev
cfile-prod-bkp-1-sd: askdir.c:395 >dird: CatReq
Job=cFile1_Full.2007-06-04_11.37.16 CreateJobMedia FirstIndex=1
LastIndex=43682 StartFile=3 EndFile=3 StartBlock=0 EndBlock=1110 Copy=0
Strip=0 MediaId=20
Damaged buffers found at bnet.c:288
discovery of data overrun.
Buffer address: 2a985a5010
Damaged buffer: 900016 bytes allocated at line 130 of
cfile-prod-bkp-1-sd block.c
04-Jun 11:42 cfile-prod-bkp-1-sd: ABORTING due to ERROR in smartall.c:430
Damaged buffer found. Called from bnet.c:288
Kaboom! bacula-sd, cfile-prod-bkp-1-sd got signal 11. Attempting traceback.
Kaboom! exepath=/usr/sbin/
cfile-prod-bkp-1-sd: signal.c:138 Working=/var/bacula
cfile-prod-bkp-1-sd: signal.c:139 btpath=/usr/sbin/btraceback
cfile-prod-bkp-1-sd: signal.c:140 exepath=/usr/sbin/bacula-sd
Calling: /usr/sbin/btraceback /usr/sbin/bacula-sd 5335
cfile-prod-bkp-1-sd: signal.c:165 Doing waitpid
With 512 KB blocks, I get around 45-50 MB/sec. This isn't bad, but I
know that tape drive can do better. Using 512 KB blocks with tar gives
me about the same speed as Bacula:
bash-3.00# tar -c --totals -b 1024 -f /dev/nst0 .
Total bytes written: 1178075136 (1.1GiB, 45MiB/s)
Is anyone else seeing this problem with large block sizes? I know these
faster tapes drives seem to like those bigger block sizes. =)
Thanks!
-Mikey
--
Mikey Orr
Core Technologies Unix System Administrator
ITS < at > University of California, Santa Cruz
Call me at (831) 459-4222
