Since November 2018 we have performance problems with Zabbix 3.4.10.
Since this time, we made tuning for tcp connections, tuning in the OS (CentOS 6.9), tuning in the MariaDB conf, tuning in the Zabbix Server conf and tuning in the Zabbix Proxies conf.

The following problem remains:

2 times a day (5:22 p.m. and 8:22 p.m., we have different hours in the past) History Syncer reaches 100% and proxies become "NoDataProcessing".
While history syncer is at 100% several "NoData" alarms start to appear.
To recover, we stop the 2 major proxies, restart the server and raise the 2 proxies with a time interval.

Number of hosts 6679
Number of items 732813
Number of triggers 358638
Required server performance, new values per second 2695.54

Hardwares (VM):

• Server:
  CPU: 4 CPU
  Memory: 24GB

• Database (Ver 15.1 Distrib 10.3.4-MariaDB, for Linux (x86_64) using readline 5.1):
  CPU: 8 CPUs
  Memory: 64GB
  Disk: SSD

Infrastructure:

1 Zabbix Server version 3.4.10 (lx-all-mon-zserver-01)
1 Database Ver 15.1 Distrib 10.3.4-MariaDB (lx-all-mon-zdb-03)
6 proxies for machines
1 proxy for Databases monitoring
1 proxy for webchecks

Attachments:

Server, proxies and database configurations.
Server and database graphics