One core gone wild

General questions
, ,
1

We’ve got VyOS 1.4.0 running on a fitlet3 with the Celeron J6413. It seemed to look good in testing so we deployed it today. It’s function is simply to do Netflow on traffic that passes through it. Routing is basic eBGP one uplink and one downlink.

It looks like the CPU is handling the Netflow load of 880 Mbps 5 minute average and peaks to 920+ Mbps better than the previous box, but we have one CPU core that not long after deploying went up to 100% and just stays there. The process that appears to be creating that load is kworker/0:2+kacpid

image
image

2

This appears to be a common problem with some hardware, Linux related not specifically vyos

Does any of the information in this link help?

3

Thanks, in the small amount of research we did I was gathering the same idea that it wasn’t likely related to VyOS specifically rather Linux in general. I’ll follow up on that link tomorrow.

4

Try sFlow instead of netfow
https://docs.vyos.io/en/latest/configuration/system/sflow.html?highlight=Sflow#configuration

5

Also try to enable all the offload options for involving interfaces and see if that does any change?

6

We’re using Netflow to send to Ubiquiti’s UISP software. I’m not sure if it supports sFlow or not. I’ll look into it. I’ve never used sFlow.

7

I think the only offloading our NIC supports is TSO, for segmentation.

8

So try them all to see if anything changes (with reboot in between)?