On Wed, Apr 07, 2010 at 08:41:41AM -0400, Il Neofita wrote:
> Hi
> I have very poor performance when the agent send the spool to the bacula
> server

First the obvious question: are the server or client doing anything
else while the backup is running?

> I currently use a RAID5 SAS 15K, 1GB Ethernet
>     hdparm -t /dev/sda2

is that hdparm run on server or client ?

> /dev/sda2:
>  Timing buffered disk reads:  692 MB in  3.00 seconds = 230.41 MB/sec

that is raw buffered linear burst read speed... seeks can easily slow
it down for two orders of magnitude or more -- and filesystems do
require seeks.

And if there are going to be lots of seeks or less depends on many
factors (like what FS are you using, the amount of file fragmentation, 
are we talking about one big file or many many small files etc).

> Any idea what I can check?

1) speed of writing on SD (server) side
   sync ; echo 3 > /proc/sys/vm/drop_caches
   time (dd if=/dev/zero of=your_spool_file bs=1k count=1000000; sync)

   # fix location of your_spool_file, and size of spool 1000000*1k
   # all examples assume that your dataset is at least 1GB large
   # you should probably use spool size at least 4 times the amount
   # of RAM to get accurate results.

2) speed of reading on FD (client) side
   sync ; echo 3 > /proc/sys/vm/drop_caches
   time tar zc / | dd of=/dev/zero bs=1k count=1000000

   # omit 'z' if you're not doing compression, change '/' to match your fileset
   # note that if you're using encryption it will slow down things considerably

3) network speed between FD and SD
   # on server
   nc -l -p 12345 > /dev/null

   # on client
   time (dd if=/dev/zero bs=1k count=1000000 | nc -w1 server_ip 12345)

   # server_ip is your server IP address

4) combination of the above, should be similar to the bacula speed

   # on server
   sync ; echo 3 > /proc/sys/vm/drop_caches
   nc -l -p 12345 > /dev/zero

   # on client
   sync ; echo 3 > /proc/sys/vm/drop_caches
   time (tar zc / | dd of=/dev/zero bs=1k count=1000000 | nc -w1 server_ip 
12345  ; sync)

those should get you some idea about what part is bottleneck...

------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Bacula-users mailing list
Bacula-users AT lists.sourceforge DOT net
https://lists.sourceforge.net/lists/listinfo/bacula-users