Gaps in graphs, poller process constant ~37% busy, server resources barely touched

[danodemano]

The plot thickens. Just for fun this morning I loaded up the appliance into my ESXi environment. I configured the two network interfaces, and added in the same 6 devices I was playing with before. And amazingly it's performing great:



The poller is still a constant ~30% busy but the graphs it's generating are perfect. I'm not sure what this means really...My previous (problematic) install was on a fresh CentOS install with the default MySQL and Zabbix configurations that I later tweaked to try and solve the issue.

I don't have high expectation for the appliance performing when I throw another 50+ devices at it but I feel like there is something to be learned here. How is this working flawless on 512MB of RAM and one CPU core when my other install with 8GB of RAM and 4 CPUs cores had massive gaps in the data? Any insight would be appreciated. Thanks!

[danodemano]

Alright folks, the appliance test was a failure as the bandwidth climbed the graphs went to garbage again. I did however find the solution and wanted to post back in the event someone finds this thread in the future.

There are two different octet counters when pulling via SNMP. There is an ifInOctets and an ifHCInOctets which are 32bit and 64bit counters respectively. For 1gbps and above interfaces where the usage is consistently high (seems to be >400mbps) the 32bit counter overflows so quickly that the data returned by it is useless. The solution is to use the 64bit counter and as you can see this produces very nice and accurate graphs:



Hopefully this helps someone in the future. Thanks all for the input!