Zabbix proxy over a high latency link

[Petr Sehnal]

Hello,
We are troubleshooting a Zabbix Proxy (v7.4.5, server 7.4.5) located over a high-latency link with approximately 250ms RTT. This proxy shows high Data Sender utilization (17% average, spiking to 80%). We have another proxy with an identical configuration and load that shows Data Sender utilization close to 0.5%, but it has a much lower RTT of 22ms.

We investigated:
MTU/Fragmentation, network path is clean (MTU adjusted), and no fragmentation is occurring.
Batching: The History Cache is consistently near empty. The proxy appears to be in "real-time" streaming mode.
TCP Overhead: Because data arrives in small increments, the proxy initiates a new TCP session for almost every individual value/small batch. With a 250ms RTT, the ~500ms is required for the TCP handshake and teardown.

Is there any way to force a Zabbix 7.x Active Proxy to be less "chatty" and buffer data into larger batches before initiating a connection? Or is this high utilization expected when operating over high-latency links?

Any best practices for high-latency tuning in Zabbix 7.x would be appreciated.

Thank you.
Petr

[guntis_liepins]

you may play with ProxyConfigFrequency and DataSenderFrequency , but you must adjust triggers accordingly.

[Markku]

Hello,
TCP Overhead: Because data arrives in small increments, the proxy initiates a new TCP session for almost every individual value/small batch. With a 250ms RTT, the ~500ms is required for the TCP handshake and teardown.

Is there any way to force a Zabbix 7.x Active Proxy to be less "chatty" and buffer data into larger batches before initiating a connection? Or is this high utilization expected when operating over high-latency links?

Any best practices for high-latency tuning in Zabbix 7.x would be appreciated.

Thank you.
Petr

You are correct that Zabbix components create a separate TCP session for each "thing", so it will take long time to make a connection with such a high latency.

You can search support.zabbix.com for an existing feature request about that.

I'm pretty sure I have seen someone make an HTTP proxy for Zabbix protocol (for transporting it over HTTP/S), but I don't recall if it also did connection pooling. If you have a serious business case for improving the protocol, it shouldn't be too hard to build a custom TCP proxy tunneling in both ends of the connection so that the high-latency part of the connection is not closed between the Zabbix component connections.