Performance questions and tuning

[motarion]

Hi all !

I got some questions about performance. For the moment my Zabbix server is functionnal but some checks which are configured to be checked each 60 seconds takes 120 seconds to be checked. I can see that in the "Latest Data" tab.

If the trigger is planned to warn us after 3 failures (.max(#3)) I should receive the message 3 minutes after the first failure. But in fact I receive it 6 minutes after.

So I think I got a performance issue.

I have:
Hosts: 458 (408/9/41)
Items: 24001 (10595/4152/9254)
Triggers: 5347 (5321/21 [40/336/4950])
Required server performance: 151,16

The load on the server is quite low, 0.51 and there is 5Go of Ram. Processor is a Xeon 3,4GHz with 4 cores.

The Database is in InnoDB with replication on a second server.

I don't really know where to start to improve performance and tune it. i would like the checks to really check on the configured interval and not the double.

If you have any info or question to help us, feel free

Thank you for your time !

PS: There is a mysql replication on another server in case the first one crashes. On this one Zabbix is shutdown to prevent double checks. There is also website that use the MySQL replication on the second server but it's not an heavy use.

[ahowell]

If your values per second is only around 150 you shouldn't have performance issues. If you do, it's most likely either DB issue, or issues with the agents not responding before the timeout.

Have a read of http://www.zabbix.com/img/zabconf201...nce_Tuning.pdf

I would definitely suggest setting up monitoring of zabbix itself. See http://blog.zabbix.com/monitoring-ho...esses-are/457/

Also look into tuning your mysql DB

[motarion]

Thank you, starting to read it.

Really interesting, I'll try the Zabbix process monitoring!

[motarion]

I just put the Zabbix Server template and received alerts.

This template is REALLY usefull.

Some processes are 100% busy:
- Pollers Processes
- Unreachable Poller Processes

I got the vanilla Zabbix Conf, so if I check it I got:

# Number of pre-forked instances of pollers
# Default value is 5
# This parameter must be between 0 and 255
StartPollers=4

# Number of pre-forked instances of pollers for unreachable hosts
# Default value is 1
# This parameter must be between 0 and 255
#StartPollersUnreachable=2

I think I should change those numbers to double them at least ? I don't want to crash the server too.

I've read the PDF file on performance and they put the StartPollers to 80.

Need advice on that.

Thank you !

[motarion]

After using this template, we tried to rise the poller number and the unreachable number.

MySQL was saturating. So we migrated the ful server on a more powerfull one.

Alerts comes really fast now ! Zabbix Queue is shorter than ever ... We didn't notice this before, our number of hosts and triggers rised too fast ...

Now it's Clean !

Thanks again for the info ! Hope this will help someone else on this matter