Am Montag, den 21.06.2010, 01:20 +0200 schrieb Lukas Kolbe: > Thanks for the hints. Changing the block sizes to the ones you mentioned > doesn't help unfortunatly. > > Maybe it is a completely different problem: The backups are made to the > diskpool in parallel, from a few different clients. The SD writes them > to disk with about 200MiB/second as a whole and from 5 to 50 MiB/second > per job. That means that a single job is fragmented in many Volumes. > > I've just activated data spooling for the copy job, and even the > spooling process doesn't get faster than 20MiB/second while still > consuming one full cpu-core. > > The volume size is 32GiB if that matters (I don't want to have too many > files in the diskpool). > > I have no idea why bacula-sd consumes so much cpu-time and obviously > limits the throughput here.
Okay, I have to correct the 20MiB/sec figure for the tape drive. After I activated Data Spooling for the copy job, I installed sysstat and saw this: 01:10:01 AM tps rtps wtps bread/s bwrtn/s 01:20:01 AM 716.75 253.16 463.59 64086.25 7226.24 01:30:02 AM 494.51 166.13 328.38 42533.10 5090.18 01:40:01 AM 308.48 142.18 166.30 36369.89 2565.75 01:50:01 AM 301.74 127.34 174.39 32536.13 2694.53 02:00:01 AM 306.21 125.27 180.94 31923.61 2791.23 02:10:01 AM 403.41 147.37 256.04 37593.61 3979.05 the sd reads the volumes from disk with 20-30MiB yet writes to the spool with only 1.5 to 4MiB/sec. This night he got a whopping 33GiB into the spool. For comparison, I dd'ed a volume to /dev/null while the copy job was running: [r...@shepherd ~]# dd if=/var/bacula/dp/fs1/Vol0070 of=/dev/null bs=1M 9175040000 bytes (9.2 GB) copied, 12.0225 seconds, 763 MB/s But dd'ing it to another file reveals a problem with the storage subsystem I believe: [r...@shepherd ~]# dd if=/var/bacula/dp/fs1/Vol0070 of=/var/bacula/dp/fs2/xxx bs=1M 849346560 bytes (849 MB) copied, 32.665 seconds, 26.0 MB/s - and dd is also consuming one full core while doing this. Both filesystems are LVM-Volumes with ext4 on the same RAID-Array on an Adaptec 52445. This just shouldn't happen. I'll report back when I find a solution to this problem, thanks for your patience and sorry for accusing bacula-sd of being at fault here! -- Lukas ------------------------------------------------------------------------------ ThinkGeek and WIRED's GeekDad team up for the Ultimate GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the lucky parental unit. See the prize list and enter to win: http://p.sf.net/sfu/thinkgeek-promo _______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
