View Full Version : Performance when monitoring a big installation with Zabbix?
Hallo
We have a “simple” monitoring comparing current state (u/d) and values (error and utilization counters) with expected states. It also checks that expected values are received (reaction on for example disappearing interfaces) and it found current values have an expected value (reaction on new but overlooked/forgotten to enable items). We compare 30.000 items with an average of 5 parameters (u/d, counters) every minute. We want to replace our old self made monitoring with a new system and wonder if Zabbix do have the performance for that task?
Gert and Gunnar
nelsonab
13-01-2009, 17:57
I feel that Zabbix can handle what you are looking to do.
There are some things to consider when you're looking at Zabbix, the biggest of which is a core methodology. At the most basic level Zabbix will not tell you a yes or no if something is running. It will however tell you the quality of how something is running which in turn you can use to determine your yes or no value for something working. If you keep this in mind when designing your monitoring solution you will find it easier to design.
If you have more questions feel free to ask away. :-)
We compare 30.000 items with an average of 5 parameters (u/d, counters) every minute.
I am not sure I fully understand this. Does it mean that you collect information from 30.000 items with the refresh rate of one minute per item? If so, it means that you must be able to process 500 new values per second, which is absolute possible with Zabbix. Consider using Proxy based approach if polling is heavily involved.
One option which we still need is the SNMP Data with Dynamic Index. What do you think about the performance by our number of items.
Gunnar
I am not sure I fully understand this. Does it mean that you collect information from 30.000 items with the refresh rate of one minute per item?
Dear Alexei,
worse: 30.000 x 5 items :D
Gunnar & Gert
rimestad
14-01-2009, 14:20
Hi again,
Test with Zabbix shows severe performance problems using dynamic indexes. The system will indeed, as stated in the manual, cause heavy load on the router monitored. We measured 100% cpu load on a cisco 2610 series monitoring 100 interfaces for ifOperStatus up. For us this is a knock out creteria, so we withdraw our question.
Positive: We like the look and feel of Zabbix very much.:)
Chers Gunnar and Gert.
worse: 30.000 x 5 items :D
Why worse, it is nice to see ZABBIX used for heavy computing! All right, it translates to 2500 of checks per second or 216M of new values per hour. This is still possible provided you have adequate hardware for the central server and you have no extreme requirements for storage of the collected data, i.e. detailed history will not be kept for years.
rimestad
14-01-2009, 15:29
The server was a poor old pc but it had no problems. The router was the limitation due to massive snmp gets. Thus we believe the server could do it, but not our routers, which are for routing and not for snmp gets. What does it cost to get the snmp dynamic index problem fixed by tomorrow:D:D:D
Gert and Gunnar
The server was a poor old pc but it had no problems. The router was the limitation due to massive snmp gets. Thus we believe the server could do it, but not our routers, which are for routing and not for snmp gets. What does it cost to get the snmp dynamic index problem fixed by tomorrow:D:D:D
A poor old pc won't handle 2.5K of new values per second for sure. You need something better, much better!
Please send a message to us if you need a quote at http://www.zabbix.com/contact.php