John
On 10/9/06, John Goerzen <
[EMAIL PROTECTED]> wrote:
Hello Bacula folks,
We received this message at the Debian bug-tracking system, and thought
that this is probably a more appropriate forum for it:
----- Forwarded message from Anders Boström <[EMAIL PROTECTED]> -----
From: Anders Boström < [EMAIL PROTECTED]>
Date: Wed, 16 Aug 2006 17:32:28 +0200 (CEST)
To: Debian Bug Tracking System <[EMAIL PROTECTED] >
Subject: bacula: backup is slow
Package: bacula
Version: 1.38.11-1
Severity: normal
We have a problem with very long backup-times of our file-server.
The file-server is only used for NFS and is quite heavily loaded. It
is a Debian etch amd64 with a 2.6.16.20 kernel running on a Athlon 64
X2 3800+ (dual core) with 2Gb mem and two 500 Gb discs in raid1. ext3
is used as filesystem, with dir_index enabled.
The backup-server is also a Debian etch amd64 but with a 2.6.15.6
kernel and an Athlon 64 3200+ CPU with 512 Mb mem. Both bacula-sd and
bacula-dir is running on the backup-server, as well as a
mysql-daemon. Discs are used for backup.
The file-server and the backup-server are connected via an GE-network
without loss.
I've done some measurements on a ~7.5 Gb directory tree with 88208
files (just one user-account):
bacula backup without SW compression: 1 hour 45 mins 2 secs
bacula backup with SW compression: 2 hours 42 mins 11 secs
local tar on the fileserver*: 53 mins 3 secs
* time /bin/sh -c "tar cf - directory | cat >/dev/null"
The CPU's on the fileserver was unloaded during all operations, one
of them was 100% idle, the other was mostly >90% IO-wait. The
backup-server is 98%-100% idle at all time.
Why is the local tar *much* faster then the bacula-fd??? And why is SW
compression making the backup much slower, even with unloaded CPU's?
I've studied the network traffic pattern with ethereal during a
backup, and the backup-server is responding directly at all time. It
is waiting for data from the fileserver. I've seen up to 400 ms
without a single packet from the fileserver and most of the packets
sent from the fileserver to the backup-server is small.
What can explain this? How can I debug/analyze this further. What
should I try?
/ Anders
-- System Information:
Debian Release: testing/unstable
APT prefers testing
APT policy: (500, 'testing'), (500, 'stable'), (50, 'unstable')
Architecture: amd64 (x86_64)
Shell: /bin/sh linked to /bin/bash
Kernel: Linux 2.6.15.6
Locale: LANG=C, LC_CTYPE=C (charmap=ISO-8859-1) (ignored: LC_ALL set to sv_SE)
Versions of packages bacula depends on:
ii bacula-client 1.38.11-1 Network backup, recovery and verif
ii bacula-server 1.38.11-1 Network backup, recovery and verif
bacula recommends no packages.
-- no debconf information
----- End forwarded message -----
----- Forwarded message from Anders Boström <[EMAIL PROTECTED]> -----
From: Anders Boström <[EMAIL PROTECTED] >
Date: Fri, 18 Aug 2006 10:15:32 +0200 (CEST)
To: [EMAIL PROTECTED], [EMAIL PROTECTED]
Subject: Re: Bug#383332: bacula: backup is slow
>>>>> "JG" == John Goerzen <[EMAIL PROTECTED]> writes:
Hi John!
JG> On Wed, Aug 16, 2006 at 05:32:28PM +0200, Anders Boström wrote:
>> We have a problem with very long backup-times of our file-server.
JG> I suspect this is not a Debian-specific problem and also probably not a
JG> bug. I would suggest that you post on the bacula-user mailing list.
JG> This is likely a configuration issue that could be related to anything
JG> such as the storage of filenames in the database, network buffer size,
JG> etc.
Yes, I agree that this probably isn't a debian specific
problem. However, I'm quite sure we can rule out the backup-server as
ethereal tells me that the backup-server responds directly to all
packets from the file-server, but the file-server sometimes don't sent
a single packet for 400 ms.
JG> I should also say that the suggestion that software compression wouldn't
JG> slow things down is incorrect. It certainly will, in any system. The
JG> only possible time that it won't is if the CPU is so much faster than
JG> the data pathways that it won't slow it down, but you don't know that
JG> from the figures you posted.
I agree that software compression should slow down the backup unless
you are communication limited. And we are not communication limited in
this case. BUT we are not CPU-limited either (one CPU 100% idle during
backup, the other mostly >90% in IO-wait). We should *only* be disc
IO-limited, and even the fastest case, local tar, is very slow. At
only 1162 kbyte/s (bacula-fd without SW-compression), the
SW-compression should not slow down the backup more than a few
percent. As a comparison, 'gzip -6' (as should be used according to
the manual) on this machine sustain ~18 Mbyte/s.
Also, why is bacula-fd without SW-compression much slower than tar?
But, as you suggested, I should try the bacula-user mailing
list as this probably isn't debian-specific.
/ Anders
----- End forwarded message -----
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Bacula-users mailing list
Bacula-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-users
--
John M. Drescher
------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
