I need tips to increase NVPS

[solution]

Hello everybody,
The required NVPS is 1068
but current maximum NVPS +830 NVPS.

Could you give me some things to change to improve and reduce the queue?

CONFIG Hardware and OS:
OS: Linux version 5.10.0-25-amd64 ([email protected]) (gcc-10 (Debian 10.2.1-6) 10.2.1 20210110, GNU ld (GNU Binutils f
OS (Short): Debian GNU/Linux 11 (bullseye)
CPU Intel(R) Xeon(R) CPU E3-1240 v3 @ 3.40GHz
Memory: 32 GB
Model: SYS-1UUP-01-AI036
Vendor: Supermicro
Disk: 4 x 500GB SSD SATA RAID 10

Below is the configuration of my Zabbix Server:

System Information:



Gráfico NVPS



MARIADB CONF- 50-server.cnf: **ONLY Values Enable

[server]

[mysqld]


pid-file                = /run/mysqld/mysqld.pid
basedir                 = /usr

skip-name-resolve    = ON

bind-address            = 127.0.0.1


key_buffer_size        = 1M
max_allowed_packet     = 32M
thread_stack           = 192K
thread_cache_size      = 8
max_connections        = 200
query-cache-size = 0
query-cache-type = 0
join_buffer_size = 32M
table_definition_cache = 700
table_open_cache = 700
symbolic-links=0
max_heap_table_size=32M
tmp_table_size=32M
innodb_flush_log_at_trx_commit=0
innodb_flush_method = O_DIRECT
aria_pagecache_buffer_size = 2M
wait_timeout = 45
interactive_timeout = 45
net_read_timeout=90
net_write_timeout=90
performance_schema= ON enable PFS
query_cache_limit = 1M
binlog_cache_size = 256M
slow_query_log = ON


general_log_file       = /var/log/mysql/mysql.log
general_log            = ON

log_error = /var/log/mysql/error.log
slow_query_log_file    = /var/log/mysql/mariadb-slow.log
long_query_time        = 2





character-set-server  = utf8mb4
collation-server      = utf8mb4_general_ci


innodb_buffer_pool_size = 16G
innodb_log_file_size = 4G

[embedded]

[mariadb]

[mariadb-10.6]
​

Zabbix Server CONF ** ONLY Values Enable

LogFile=/var/log/zabbix/zabbix_server.log
LogFileSize=10
PidFile=/run/zabbix/zabbix_server.pid
SocketDir=/run/zabbix
DBHost=localhost
DBName=zabbix
DBUser=zabbix
DBPassword=123test
DBSocket=/var/run/mysqld/mysqld.sock
StartPollers=50
StartIPMIPollers=0  
StartPreprocessors=20
StartPollersUnreachable=80
StartHistoryPollers=25
StartTrappers=90
StartPingers=10
StartDiscoverers=15
SNMPTrapperFile=/var/log/snmptrap/snmptrap.log
HousekeepingFrequency=20
MaxHousekeeperDelete=1000000
CacheSize=2G
StartDBSyncers=25
HistoryCacheSize=512M
HistoryIndexCacheSize=512M
TrendCacheSize=512M
ValueCacheSize=2G
Timeout=30
TrapperTimeout=30
FpingLocation=/usr/bin/fping
Fping6Location=/usr/bin/fping6
LogSlowQueries=3000
StatsAllowedIP=127.0.0.1
​

Thank you for your support

Wellington

[cyber]

These numbers are not exactly the same... one is calculated based on your item count and polling intervals, other is the real throughput.. Adding CPU or memory does not necessarily change anything.
The other keyword what you mention is "queues". That is more important... Having high load and cpu utilization on host itself and if you have high utilization of different processes and you actually see queues, where items cannot be polled fast enough, then adding different pollers might help. If it is a all-in-one host, then splitting it up to server (and web) and DB might help free up resources.. DB performance is most important in Zabbix, if it cannot handle things, everything slows down...

[solution]

This server is for Zabbix only
zabbix server 6.4.7 + zabbix web (max 2 simultaneous users) + MariaDB 10.6

So NVPS depends on Database performance?
Or does it depend on Database + Pollers/Trappers?

there are 1500 hosts, 98% Zabbix Active Mode.
Templates: default for Windows (service discovery disabled)

Below are graphs from the last 7 days:

CPU



Memory:


Disk


Zabbix:Value Cache


Zabbix




Wellington

[cyber]

So NVPS depends on Database performance?
Or does it depend on Database + Pollers/Trappers?

It depends on how often and how many polls you do ... same amount of items but 1m vs 5m intervals is big diff in NVPS..)

https://www.zabbix.com/documentation...atus_of_zabbix

Required server performance, new values per second

The expected number of new values processed by Zabbix server per second is displayed.

Required server performance is an estimate and can be useful as a guideline. For precise numbers of values processed, use the zabbix[wcache,values,all] internal item. Enabled items from monitored hosts are included in the calculation. Log items are counted as one value per item update interval. Regular interval values are counted; flexible and scheduling interval values are not. The calculation is not adjusted during a "nodata" maintenance period. Trapper items are not counted.

These graphs look just fine to me... server resources do not seem to be overutilized.
What pops out from here is, that during weekends load (and also DB queries) is much lower, than on workdays. I doubt you have some different polling schedules on weekends, so I think its the users, who drive the DB usage on workdays... You can even tell, that they probably have regular lunches..

But then again .. load between 01 and 0.4... not really an issue...

All in all, I would would not worry about this .. It is normal to have difference in estimate and real ..

[solution]

Thank you very much for the comments
I will continue monitoring NVPS and hardware usage.

...What pops out from here is, that during weekends load (and also DB queries) is much lower...

Sorry, I forgot to inform you that I use the maintenance period without data collection.
PC Users: maintenance without data collection from Monday to Friday between 6:00PM and 7:00AM and full day weekends.
Servers: Monitoring 24/7/365

as for the tip above the key
I created it just to collect for 12 hours




Thank you very much
Wellington