Cacti maximum metric count test

Cacti has two polling methods: cmd and spine. According to the official introduction, spine polling is much faster than cmd.

Host Configuration, 4-Core 2 GB memory for KVM virtual machines

42 monitored hosts, 2070 data queries, and 1399 RRDsProcessed

CMD tuning log

Obviously, 32 processes are concurrent and have the best performance, <12 s. An average of about 4 hosts takes 1 second, and the monitoring load does not exceed 1

Spine tuning log

The total number of threads is 4*8, and the performance is the best. <5 s, it takes 1 s for about 10 hosts on average, and the monitoring load does not exceed 1.

In conclusion, the following conclusions are drawn:

Theoretical maximum value	1	5	Threrld
Cmd	42*60/12 = 210	1050	32*1
Spine	42*60/504	2520	4*8

In actual deployment, if spine polling is used, a 4-core, 4-thread, 2 GB memory virtual machine will monitor the system based on the down-server duplication and error halving standard budget. If the system collects data every minute, it can monitor 250 servers, set up to monitor 1250 sets every 5 minutes.

The test time is short. The actual environment is limited by hardware and network configuration, and whether other services are running on this server. For example, when collecting data, I ran a backup script, the collection time is more than 10 s, and the time has increased more than twice. Therefore, the above data is for reference only!

Cacti details: click here
Cacti: click here

In RHEL6.4, the Cacti + Spine monitoring host is used to send mail alarms.

Use Cacti + Spine to monitor remote hosts in RHEL6.4

CentOS 5.5 complete installation of Cacti + Spine

Cacti construction document under CentOS 6

Detailed description of Cacti monitoring deployment under RHEL5.9

How to install Cacti in CentOS 6.3

Quick installation and configuration of Cacti in CentOS Linux