[ixpmanager] SFLOW Under Reporting?

[Ian Chilton]

Hi Nick,

> for reference, inex's cache flushes take ~2-3 seconds. It may be happening that you're i/o bound on your RRD spool disk.

That’s very interesting!! - over a tenth of ours!

What hardware / virtual machine are you running that on?

What OS / version of perl are you using?

When I’ve looked previously, our traffic levels have always been very similar to yours.

For the new deployment, I’m using a VM with 4x VCPUs, 4G RAM.

/data (XFS) and /var/lib/rrdcached (ext4) are dedicated mounts and the only disks on a dedicated NVME drive.

The original deployment’s disks are one of many on another nVME.

When I’ve looked previously, we’re under 25% CPU usage, plenty of free RAM and iowait was between 1% and 3%.

I’m not seeing an improvement on the new deployment, or if I migrate the VM around, or when traffic is lower, or when I increase/reduce the sample rate.

This all leads me to the conclusion that it’s not resource or I/O bound.

Away from the keyboard at the moment, but I will look again in light of this.

Thanks,

Ian