[Bacula-users] Low performance for Proxmox VM backup

Hello,

I use Bacula daily and faced no real issues for years.

I installed a new instance and am facing a strange issue.
I installed a new bacula server under Debian Wheezy.
I try and backup several virtual machines hosted on 2 proxmox servers. 
These proxmox servers use their local hard disks drive (software RAID 1)
Most of these VM are quite small and performance is not an issue.
For 2 of them, it's not the same deal. They have 200 GB data pieces to 
back up :
  - one mail server
  - one web server

Example :

2013-12-08 17:39:51   bck-dir-01-dir JobId 80: Bacula bck-dir-01-dir 5.2.6 
(21Feb12):
   Job:                    Backup_eagle-vmail_QG.2013-12-08_10.13.18_38
   Backup Level:           Full (upgraded from Incremental)
   Client:                 "eagle-fd" 5.0.1 (24Feb10) 
i486-pc-linux-gnu,ubuntu,10.04
   FileSet:                "VMail q-g" 2013-12-08 10:13:18
   Pool:                   "File" (From Job resource)
   Catalog:                "Catalog1" (From Client resource)
   Storage:                "File" (From Job resource)
   Scheduled time:         08-déc.-2013 10:13:17
   Start time:             08-déc.-2013 10:13:20
   End time:               08-déc.-2013 17:39:50
   Elapsed time:           7 hours 26 mins 30 secs
   Priority:               15
   FD Files Written:       426,533
   SD Files Written:       426,533
   FD Bytes Written:       79,375,820,932 (79.37 GB)
   SD Bytes Written:       79,463,703,342 (79.46 GB)
   Rate:                   2962.9 KB/s
   Software Compression:   None
   VSS:                    no
   Encryption:             no
   Accurate:               yes

All machines have their network link set up to 1Gb/s FD
With SCP, I can send / receive files at 50-60MB/s.

First I used one job. It was running at something around 3MB/s.
I split this job into 5 pieces. The job you can see backs up all mailboxes 
beginning with Q, S, D, F, G.
Each job runs at 3MB/s. So, with 5 jobs running at the same time, I can get 15 
MB/s. The backup window has been reduced by 5 and is now acceptable.

I disabled "Accurate=yes". no change in the job's duration.
I enabled compression : no change in the job's duration.

It's not bound by the CPU neither the network. (running more jobs just 
increases the throughput linearly)

With 1 job running at a time, there is no iowait.
With 3 jobs at the same time, I have less than 5% iowait

What could I check to find where is the bottleneck ?

Best regards.
Jerome Blion.




------------------------------------------------------------------------------
Rapidly troubleshoot problems before they affect your business. Most IT 
organizations don't have a clear picture of how application performance 
affects their revenue. With AppDynamics, you get 100% visibility into your 
Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro!
http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users