How to resolve xenbr0 packet drops in XCP-NG

Computer Hardware & Server Infrastructure Builds
1

I noticed packet drops in dom0 on xenbr0 when under load. Not lots, say under 1% and not related to vlan tagging.

My first thought was to increase the ring buffer, but ethtool says this is an openvswitch bridge and that’s not something it can do. Changes to eth0 have no effect on xenbr0, and moving traffic to another card just moves the dropped packets to the new xenbr3.

What’s the next step?

in reading on on that I saw some mentions that you can have a condition where the flow-eviction-threshold is exceeded, but no leads on how to check that.

2

Not an issue I have run into but I have had poor performance when using Broadcom cards. My solution was to move to Intel based cards.

3

I’d read the same thing, but I’m using intel cards. Same dropped packets on both the built-in Intel I219-V and the converged x540 I added. I think ovs-ofctl show xenbr0 shows the buffers set to “n_buffers:256”

Quite the mystery.

4

I moved the workloads over to another pool member (a dell 3070 with integrated nic) and it has no issues. So I have a feeling it’s specific to the H270 hardware/kernel mix. If I get motivated to test a newer kernel I’ll update.