Backup Central

Dear All

I have Bacula-2.0.3 running on a dual P3 1 GHz linux system with a
2.6.21.1 kernel.

I have noticed that the backup rate of a client decreasing during the
backup it starts at 3,200 KB/s and slows down to 2,000 KB/s after 30
minutes. I used Status of the Storage daemon to get the rate about every 5
minutes. The backup was 6 GB to disc.

I suspect the slow down may be due to the fact other jobs were waiting for
this job to finish since I ran another backup of another client and the
rate was constant during the whole backup, the difference being there were
no jobs queued to run ie waiting for the device.

Could the slow down be caused by the director polling the system and so
impacting on the storage daemon?

FYI I had a swapper error reported by the kernel and thought it was the
problem - I had forgot to tweak the TCP settings on the "new" system. The
above backups were done with the tweaks applied.

The tweaks are

echo "65536" > /proc/sys/vm/min_free_kbytes
echo "6553600" >/proc/sys/net/core/wmem_max
echo "8738000" > /proc/sys/net/core/rmem_max
echo "8192 873800 8738000" > /proc/sys/net/ipv4/tcp_rmem
echo "4096 655360 6553600" > /proc/sys/net/ipv4/tcp_wmem

Thanks in advance
Stephen Carr

In response to "Support" <support < at > ic...>:

I have Bacula-2.0.3 running on a dual P3 1 GHz linux system with a
2.6.21.1 kernel.

I have noticed that the backup rate of a client decreasing during the
backup it starts at 3,200 KB/s and slows down to 2,000 KB/s after 30
minutes. I used Status of the Storage daemon to get the rate about every 5
minutes. The backup was 6 GB to disc.

I suspect the slow down may be due to the fact other jobs were waiting for
this job to finish since I ran another backup of another client and the
rate was constant during the whole backup, the difference being there were
no jobs queued to run ie waiting for the device.

No. Queued jobs do nothing to affect the speed of a backup.

Could the slow down be caused by the director polling the system and so
impacting on the storage daemon?

No.

The "problem" is almost certainly caused by a change in the makeup of the
data being backed up. Since "status client" gives you the speed at the
time you run it, that can change over time if any factors change the speed.
This could be something like activity on the client contending with Bacula,
or network traffic.

However, it's most likely (in my experience) that you start out backing up
a lot of big files, then the backup moves into directories where a lot of
small files are found. 2G composed of 2 1-G files will back up faster
than if it's composed of 2000 1-meg files. This is because each file has
some overhead associated with writing backup records to the database.

If the speed difference is enough to be of concern, you should look at
optimizing whatever database backend you're using to speed up inserts.

--
Bill Moran
http://www.potentialtech.com