Constant Input discards on monitored Broadcom NIC's

[dadu]

CMK version: Enterprise 1.6.0p19

Error message: constant flow of in-discards on every interface that is a Broadcom NIC

Hi!

In the environment where I am working, we see a constant stream of in-discards on every interface that is a Broadcom NetExtreme II BCM57810 NIC.

On the host from where this information is pulled by CheckMK, these errors are not visible in the netstat, so it really looks like it is a CheckMK monitoring thingy.

Does this ring a bell anyone? Really looking for a solution on this

[andreas-doehler (Andreas)]

The information about dropped/discarded packages are coming from the output you get with.

cat /proc/net/dev

You can also check what you see with an ip -s link there should be the same data visible as from /proc/net/dev.

[dadu]

Hi Thanks for the reply!

I actually reviewed both but I cannot find the in-discards in either of those ways, but what I am seeing is that the in-discards are present on Intel cards aswell - So that is dealth with.

Now what I do see is errors whenever we Vmotion a machine, but just on the Broadcom NIC and that when I click on an interface (Broadcom / Intel) in CheckMK, I can see that the same check is used (check_mk-lnx_if), but that there are different parameters used to calculate the errors - Is that correct? can I alter this? I want to use the same calculations for both interfaces, that might reduce the errors we regularly see when we for example move virtual machines from one location to another.

Blue is the Intel NIC (where I never see any errors)
Red is the Broadcom NIC (where I regularly see errors)

Thanks so far!

[andreas-doehler (Andreas)]

The Intel NIC (blue) uses the default settings and the Broadcom has already altered settings.
But now it gets a little bit confusing. In your first post the problem where discarded packages and now the problem is in-errors?
In your screenshot you see that booth interfaces have actual discards and this is “normal” if it is not too much.
The in-errors you should have very seldom.
That’s also the reason why the default value is very low with 0.01% of packages.

Your actual rules you see if you use the

[dadu]

Yes, I had 2 so called issues, one was the in-discards, which I first thought only appeared on the Broadcom NIC’s the second issue was the in-errors that periodically appear. So the first issue is addressed as both the Intel as Broadcom cards have the same behavior, and now I am focussing on the in-errors!

Thanks for the information!