[ovs-discuss] Network get unstable, put the whole system on halt

[Jesse Gross]

On Mon, May 20, 2013 at 3:00 AM, Salman Toor <salman.toor at helsinki.fi> wrote:
> Hi,
>
> We are using openvswitch with Openstack Quantum module. Following are the
> details of our system..
>
> Openstack Controller and Compute nodes are running with Ubuntu 12.04.5,
> kernel 3.5 and OpenVSwitch version 1.4.0+build0 with GRE tunnels.
>
> The problem is with very little activity everything works very fine but as
> we started to increase the load on the system the kernel log started to grow
> on the controller and fill the entire disk space and halt the complete
> system. And it happen within 2 to 3 hours ...

This is caused by an overzealous check in the upstream kernel when
used with certain drivers and certain offloads. It's already been
relaxed upstream (in 3.9 I believe). You should be able to avoid it by
turning off GRO and/or LRO in the receive NIC, although that itself
will likely reduce performance.