Big implementation - Zabbix scaling

[DonPook]

Hello,

We are researching our options for monitoring a large infrastructure.
Devices are around 60.000 with 55k of them being IP Phones.
The thing is that we want to measure performance data (RTA) and get SLA reports based on their availability, so we would need to actually ping each IP Phone separately.
The question is, what kind of resources would I need for such a project? How many checks can 1 proxy perform without any performance issues? (Trying to figure out how many proxy servers would I need and their HW requirements basically)


Thanks in advance.

[Zdenek_OMNISENSUIT]

Hello.

As you mentioned 55k of IP phones, I think that they are in different locations, right?
If yes, you will need at lease one proxy at each location. (at least two better for HA solution)
Every proxy can handle number of tasks (NVPS) depends on resources like CPU, RAM, IOPS (SSD drive) and good setup of database under the app. (separate FS for db data, logs, enought of RAM etc).
Anyway if you will setup just ping and just for 5min each check than you will no need so big resources.
If the monitoring will be more sophisticated (50 items per host and very often) than you will need little bit more.
You can start with 4vCPU/CPU (vm/hw) and 16GB of ram per proxies and server.
(for proxies can be less depends on real number of items measured with that proxy but 2vCPU/4RAM should be the minimum for you)
This can handle a lot of checks - 50k items/400NVPS very easy with good setup and with fine CPU usage with this resources.
So, you can start with this and will see.
Than you can tune it or add resources if need.
On zabbix server please separate db and app server.
If you will need please ask.


-------------------------
omnisensuit.com

[DonPook]

Hello and thanks for your answer,

As far as I know, the IP Phones are scattered all around the country in multiple sites so I don't know whether installing a proxy in each site would be a good idea for 2 reasons:

1) Not possible to install 1 proxy per site.
2) No real RTA (The measurements should be from a local place towards the phones)

That's why I was wondering in how many proxies I need to break down to, not per site, but per max checks on a proxy.

[Zdenek_OMNISENSUIT]

Hello.

You can just use (for example):
zabbix server (separate standalone setup or HA solution) + 1proxy (2 in HA) for monitoring your phones + 1proxy (2 in HA) for monitoring the rest of your hosts.
Resource needs depends on NVPS (number of items, frequency of checks), some numbers you can count with see above.
But this can be tricky with networking (I don't know your architecrure/security etc).
Maybe you can think about do it by different way.
As you have this number of phones, maybe you have some tool for management.
Maybe there is some kind of availability check.
Maybe you can just check this (availability, errors, log, etc) and sent info to zabbix.


-------------------------
omnisensuit.com

[Hamardaban]

The question of calculating the power of the server and related proxies does not have a simple answer.
You can start operating with one proxy (or no proxy at all. only the server) .
Further, if a gradual increase in the number of monitoring objects reveals problems, you can deploy the proxy if necessary.

Tip1: Zabbix is based on a database. Everything depends on its performance. Choose the database that you know how to configure.

Tip2: if the project is really big, I strongly recommend that you take training in zabbix or study the documentation very well.

https://www.zabbix.com/documentation/current/manual/installation/requirements

https://www.zabbix.com/documentation/current/manual/appendix/performance_tuning