Zabbix CPU Overflow on the server

[olivier_cisco]

Hello guys,

I have a little question about zabbix. We plan to implement the zabbix solution as a supervision solution for around 3000 hosts.

We tried to implement on different kind of server as specified in the zabbix server specifications. Our last server type is a Linux unbuntu 64 bits with Intel Xeon 3,2Ghz with 3Gb memory.

I must tell you that the host's database is on the same physical server than the zabbix server.


We have a problem dealing with because as soon as we add our hosts in zabbix database, any request we make in the zabbix console (like refresh for example) take a very long time to be displayed on the screen.

When we look at the top command on our server, we have a cpu overflow during around 2 minutes.

Does anyone have an idea of what whe can implement on our server to increase the response time ?


Thanks ou for your responses and please do not take care about my bad english, Olivier from France.

[vinny]

Olivier,
you didn't tell what kind of DB u used : postgres, mysql (innodb) ??

Then 3 Gb RAM is clearly not enough for a zabbix server with a lot of hosts.

How many items u planned to check per host ?
Because u d better thinking of using zabbix in a cluster mode, with one master & several child nodes.

vinny

[olivier_cisco]

Hello vinny,

Sure i forgot to tell ya about the DB which is a MSQL INNODB.

We where thinking that we lack of memory on the server but i don't understand how it can lead to a cpu overflow. Maybe you have an explanation because we don't have an expensive memory use at this time.

For the question about the items, we plan to check around 3 items per host. But, at this time we have only add the hosts in the database without any item for any host. We only are using the map option and we already have cpu problem.

Our final goal is to have a map of the company network with supervision of some parameters for our 3000 cisco devices.


You said :

"Because u d better thinking of using zabbix in a cluster mode, with one master & several child nodes."

I don't understand what you are meaning ? Can you tell me more please ?

[vinny]

With ur configuration, it is weird to have a high cpu load....
which process is loading ur server ? mysql, apache, zabbix ??

For the cluster mode of zabbix, u can read the manual....there is an architecture with several zabbix server, synchronizing, etc etc

[olivier_cisco]

Vinny,

which process is loading ur server ? apache => httpd

Can i use Zabbix whitout cluster mode to supervise over 3000 hosts ?

[vinny]

clearly not possible....
maybe u can try to install the db on a separate server.


but if it is apache process that taking high cpu load, u may try a php cache & acceleration module, like xcache or eaccelerator.
that will decrease the load on http traffic.

[cstackpole]

I have a similar setup and had similar problems. Xeon 3 Ghz monitoring ~30 hosts with about 70 items per host. 90% of the items had less then 30second update intervals. I added all of the systems within a day of work and within the week my server had a load of about 10 and had many problems.

In my case, it was MySQL that was just draining the resources on the server. In the end I made the server 64bit Debian (moving from 32 to 64 helped quite a bit) then I dropped 8GB of memory into the system and spent over a month tweaking MySQL.

I don't know if this is your problem but if it turns out to be MySQL let me know and I will do my best to help.

[Edit] missed the part about you already on 64bit. Thats good. Also, in my case the 8GB was needed so I could load the majority of the database into ram. That helped me more then almost anything else.

[Alexei]

There are some issues on MySQL (InnoDB) side which makes MySQL scale in a bad way for Quad/Eight Core setups in case of a large (well, more than 8 or something) number of simultaneous transactions. You might be interested in upgrading MySQL server.