Sizing information of Cacti server

[soumendra]

Hi,

I want to deploy cacti in my production environment having 10000 Linux and 700 AIX servers with VIOS and IBM power hardware. What would be the capacity of Cacti server (CPU/RAM/Database settings) to poll data from this huge list of servers. I will generate around 10 performance graph per server using SNMP queries and will retain for two years for historical data analysis. Can someone recommend me?

[bmfmancini]

Hey there

one of my largest installs has something close to this

close to 15k devices total spread out over 3 servers

Here are the specs

Main server
Spine poller
Mariadb
16G Ram
16Vcpu

1TB storage

Poller 1
Mariadb
8Vcpu
8G ram
100G disk

poller2
Mariadb
8Vcpu
8Gb ram
100G disk


The rest is up to your network make sure you have low latency connections to the devices and make sure your network is setup for that much UDP traffic things like firewalls will need to be configured to allow that traffic without freaking out that there is a UDP attack happening also take note of your PPS as you start ramping up some older routing gear might have a hard time with a large install like that


polling times for this setup stand at around 20seconds per poller ( 1 Minute polling Cycle ) polling between 4 - 10 graphs per device
All routers and switches in my case

DB settings pretty much what Cacti suggests during the install
Linux based installs

[TheWitness]

Sean,

Post your "SYSTEM STATS:" for the main Data Collector and the Remote...

[bmfmancini]

Here you go Larry

Main
2020/07/22 10:30:26 - SYSTEM STATS: Time:23.9207 Method:spine Processes:10 Threads:10 Hosts:2327 HostsPerProcess:233 DataSources:44472 RRDsProcessed:0

Remote 1

2020/07/22 10:31:21 - SYSTEM STATS: Time:19.1262 Method:spine Processes:25 Threads:10 Hosts:6025 HostsPerProcess:241 DataSources:83409 RRDsProcessed:0

Remote 2

2020/07/22 10:32:25 - SYSTEM STATS: Time:23.1096 Method:spine Processes:20 Threads:4 Hosts:2922 HostsPerProcess:147 DataSources:43901 RRDsProcessed:0

[bmfmancini]

My proccess and Thread counts are on the high side I know
we had some network issues before so I push more threads and proc counts to accomodate
but the servers dont really feel it too much I am so impressed by the performance of spine !