How many devices is zabbix able to manage?

[txalin]

Hi all,

I'm working as a UNIX administrator in an spanish supermarket company, we have a monitoring proyect for this year and now we are evaluating a few open source solutions, at this moment zabbix is the first one in our list, we have setup an small pilot monitoring only 20 devices, and the issue that we found is that the database size is 130 mg ony in two weeks!!

So, the facts says that we need a huge server for this proyect, probably we will deploy a server attacking to a SAN were the database is located.

As you can see this is a very huge proyect, but, after starting with it we need to answer just two easy questions

We will have in the end of the proyect more than 10000 devices monitored by zabbix, using snmp and zabbix agent. Is zabbix able to handle this number of devices? I saw on the main zabbix web that it ave been tested with 5000 devices, unfortunatelly is a very small number for us...

And the othe one is..., is there any kind of scalability on zabbix servers? i mean, can we deploy some kind of collectors and a central servers like, for example, tivoli does?

Thanks for help, and keep up the good work!!

[Alexei]

The number of devices is not that important. ZABBIX performance is mostly depends on a number of values that have to be processed every second.

ZABBIX scales well up-to 800-900 of values per second, which basically means that you may monitor 60K of parameters with refresh rate of 60 seconds by a single (yet, powerful) ZABBIX server.

These numbers can be achieved using active checks. Polling requires more CPU and IO resources.

[Petya]

I've managed to achieve 6000-7000 items per second
without any triggers on a double core CPU, 16 GB RAM, InnoDB.

In a few days I'm going to test configurations with triggers.

[txalin]

Ok, so, until i understand, if i scale the checks over the time (i mean, check the 300 most important servers on first 20 seconds, so that means 5000 values, then check other 400 machines on next 15/20 seconds, with less checks, and so on...) will zabbix be able to handle this amount of devices?

Well, and regarding the database (maybe is not the appropiate forum to ask this..), we have three engines, oracle entreprise, mysql and postrge SQL, i think the best one is oracle, wich one do you think is the best one?

[Alexei]

I've managed to achieve 6000-7000 items per second
without any triggers on a double core CPU, 16 GB RAM, InnoDB.

Impressive! Use of triggers will surely decrease the performance, I think you will still be able to get 2-4K on your hardware.

[Alexei]

Ok, so, until i understand, if i scale the checks over the time (i mean, check the 300 most important servers on first 20 seconds, so that means 5000 values, then check other 400 machines on next 15/20 seconds, with less checks, and so on...) will zabbix be able to handle this amount of devices?

You are absolutely right.

Well, and regarding the database (maybe is not the appropiate forum to ask this..), we have three engines, oracle entreprise, mysql and postrge SQL, i think the best one is oracle, wich one do you think is the best one?

Choose MySQL if you are looking for the best performance!

[Petya]

I don't think Zabbix currently allows prioritization or any other
regulation of its processing queue. Though it should!

So you can't configure that some items are collected first, then other.
You have to be sure that your server is able to process
whole bunch if items.

To estimate the number of items per second you currently have,
you can use this patch:

http://www.zabbix.com/forum/showthread.php?t=8714

Take in account that I've achieved 6634 items per second
of "Zabbix Agent (active)" items, which is not possible
if you're doing SNMP polling.

[txalin]

yep, i noticed that Petya.... was thinking about how can i set it up, but no ideas come to my mind.... you know, it''s friday :P

The only thing i can "imagine" now is to setup an "strange way", but now, i need help with "Update Interval" parameter, if i setup Update interval to, for example, 20 seconds, it means that every 20 seconds the data will be inserted on database or it means that on that range of time the data will be inserted when it arrives to the servers?

I mean, Server receives size of /var partition, and we setup 20 seconds on update interval, that data will be collected every 20 seconds or will be collected every second during 20 seconds and then wait for 20 seconds until it start to collect again?

Apologize if it's a little bit "twisted", but, you know, it's friday...

[Alexei]

I mean, Server receives size of /var partition, and we setup 20 seconds on update interval, that data will be collected every 20 seconds or will be collected every second during 20 seconds and then wait for 20 seconds until it start to collect again?

Apologize if it's a little bit "twisted", but, you know, it's friday...

I wasn't sure it is Friday today until I saw your post! The data will be collected every 20 seconds, i.e. one value per 20 seconds (i.e. 180 values per hour, etc).

[txalin]

I wasn't sure it is Friday today until I saw your post! The data will be collected every 20 seconds, i.e. one value per 20 seconds (i.e. 180 values per hour, etc).

Argh!, i was thinking on scale it trough time, but no way, i can't do it.....

Well, will think on that next week

[cristhiano]

I think the core of zabbix can monitor this number. But not want to view the web interface.
For this number of hosts/itens is very slow and not friendly.
Web interface of Nagios is better.

[txalin]

I think the core of zabbix can monitor this number. But not want to view the web interface.
For this number of hosts/itens is very slow and not friendly.
Web interface of Nagios is better.

mmm... why? If fact, zabbix frontend is just a php web interface, my experience says that if it's slow will be due to the database, ins't it?

[cristhiano]

With 10.000 hosts, you'll see status for all. You can ?
If 0,5 kbytes for host, your browser download 5 Mbytes.
If refresh is 30s, never finish.

I'm have Nagios and overview interfaces is good (have resumes). You go in clicks for detail. In Zabbix appear all for all.

I am studying the zabbix and liked him. I have 4.000 hosts. And change interface for use.

Alexei say:
The overview screen is not very useful for 4K of hosts anyway...

http://www.zabbix.com/forum/showthread.php?t=8375

[txalin]

With 10.000 hosts, you'll see status for all. You can ?
If 0,5 kbytes for host, your browser download 5 Mbytes.
If refresh is 30s, never finish.

I'm have Nagios and overview interfaces is good (have resumes). You go in clicks for detail. In Zabbix appear all for all.

I am studying the zabbix and liked him. I have 4.000 hosts. And change interface for use.

Alexei say:
The overview screen is not very useful for 4K of hosts anyway...
http://www.zabbix.com/forum/showthread.php?t=8375

ok, i understand now.... will take a look, anyway i didn't like nagios interface, it's a little bit.... "ugly"

Anyway, maybe we can modify the ovierview screen, or even delete it from the panel... don't know