I've seen posts regarding attribute inserts as a performance issue even when
modern 10K 15K drives are being used. I backup 800 clients every night. It's
a mix of OS X and Windows. 90% of which is Windows XP. I run concurrent jobs
and I have five storage servers each with ~800 GB of storage. These are hand
me down files servers running PATA disks. Practically antiques! I spool all
my jobs to a single 200 GB SATA drive on each storage server and then
despool to and 800 GB PATA RAID 5. I get about 50MB/sec despool rates.
I've semi-randomized when fulls run across a 7 day period and each client
gets a full backup once a week. These fulls are weighted towards the end of
the week and a larger percentage of fulls run on Friday-Sunday.
Even though I could get faster spool/despool times with newer SAS or SATA
drives I noticed that database inserts always seem to slow the entire
process down. I assumed this was due to a lot a small writes to the InnoDB
tablespace during the batch insert process.
I tested several different drives at my disposal by importing a dump of the
catalog database. Keep in mind that maximizing write speed was my primary
goal. I tested a 10K WD Velociraptor, a Seagate 15K SAS drive and an Intel
X25 extreme SSD. This was a single drive test without any RAID just to get
an idea of the difference in drives. The SSD gave the best overall
read/write performance.
I set up a Mysql database using InnoDb tablespace with two 32GB X25 drive on
software RAID1 with hot spare. The computer has an AMD Athlon 3700+ with 4GB
of ram. I've given 1.5 GB to the InnoDB tablespace.
Prior to moving my database to SSD's my mix of nightly backups used to start
at 18:30 and the last of the jobs, depending upon the mix of Fulls and
Incrementals, would not finish until 12-16 hours later. Prior to this change
I had played with the schedules so the nightly batch would start at
18:30,22:30,and 23:55. I would never see the first batch of jobs that
started at 18:30 complete by the time the next scheduled batch would start.
Now I'm seeing all of the jobs finish with gaps of idle time before the next
schedule batch of jobs.
This change has cut my attribute insert time down to the point where I don't
consider it an issue.
Brian Kelly
smime.p7s
Description: S/MIME cryptographic signature
------------------------------------------------------------------------------
This SF.net email is sponsored by:
High Quality Requirements in a Collaborative Environment.
Download a free trial of Rational Requirements Composer Now!
http://p.sf.net/sfu/www-ibm-com _______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users
| 