Changing Active Data Gathering (moving to Proxy)

[Clontarf[X]]

Hi Friends,

I have a few Zabbix instances, all of which hover around 1k-1.5k NVPS (depending on the amount of templating I want to do).

An old post on the forums suggested that once you start reaching the "Large Environment" deployment scale, that you should shift data collection and polling to a proxy. I like this idea, as I understand Zabbix needs most of it's available resources to do the maths.

The server at the moment is virtualised, 16 cores of Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz and 32gb RAM. One of the issues I have is not seeing any real improvement in poller utilisation when I give it more cores. Is the performance limitation/utilisation coming from a lack of DB power, perhaps?

Would simply spinning up another proxy and pointing all hosts to it provide a sufficient enough performance increase/reduction in resource requirements? Most of the monitoring is SNMP so most of our resource requirement is in the pollers. This makes me think that switching to a proxy for data won't really help performance as it's mostly poller/passive monitoring. Specs below;

# cat /etc/zabbix/zabbix_server.conf | grep -e '^Start.*=.*$'
StartPollers=500
StartIPMIPollers=0
StartPreprocessors=13
StartPollersUnreachable=100
StartTrappers=20
StartPingers=20
StartDiscoverers=100
StartHTTPPollers=5
StartAlerters=20
StartVMwareCollectors=1
StartSNMPTrapper=1
StartDBSyncers=16

20mb/sec average bandwidth (most of this might be SQL traffic, haven't dug into it very far).

[cyber]

For server resources you can get away with much less, if you go with proxies...Then again, you have to spend some of resources then to proxies...

I have 4 times as much hosts, double of those triggers, similar amount of items and a bit more than twice of nvps... and server runs on 4cpu-s, 8G of mem... almost idling... load average: 0.56, 0.53, 0.55. But I have 16 proxies..

[Clontarf[X]]

Originally my deployment plan was a proxy at each location so that I could gather information during WAN outages, but it wasn't suitable due to local resource requirements and the management overhead (currently over 90 locations).

I even wrote the automation to deploy them all automatically but I am just a one man army, and 90+ proxies is just too much for me.

At least now you've confirmed that I am nearing the scale at which proxies will be needed even just to offload collection responsibility from the server.

[AsevaPliuha]

Great question! It sounds like you're running into some performance limitations with your Zabbix instances. One solution to consider is using a proxy for data collection and polling, as suggested in the old forum post you mentioned. This can help offload some of the resource requirements from the main server. However, it's also worth considering if the limitation is coming from a lack of DB power.

[AsevaPliuha]

Great question! It sounds like you're running into some performance limitations with your Zabbix instances. One solution to consider is using a proxy for data collection and polling, as suggested in the old forum post you mentioned. This can help offload some of the resource requirements from the main server. However, it's also worth considering if the limitation is coming from a lack of DB power.

One thing you can try, if you haven't already, is spinning up another proxy and pointing all hosts to it. This could provide a significant performance increase and reduction in resource requirements, especially if most of your monitoring is done via SNMP. Another solution you can look into is using fast residential proxies to speed up the data collection process.

[LenR]

What are most of your items ? I switched to Agent Active for most items.