New Question

Revision history [back]

click to hide/show revision 1
initial version

Openvswitch vxlan/geneve performance

Hi.

Using openstack pike - hyper-v on Win 2k16 - 4 node cluster - with OVS 2.7.4 for vxlan (or geneve) tunnels via neutron-ovs plugin, I am getting maximum 3 Gbits throughput between vms on separate hosts (using set switch 2x 10gb NICs and ovs bond-slb on top). That is quite disappointing. Is there any vxlan offloading mechanism possible? I've found information that it is planned on wiki of yours:

https://wiki.cloudbase.it/ovs-hyperv-architecture

I've tested disabling vmq and largesend offload as suggested in the same wiki. Not so much of a difference

Openvswitch vxlan/geneve performance

Hi.

Using openstack pike - hyper-v on Win 2k16 - 4 node cluster - with OVS 2.7.4 for vxlan (or geneve) tunnels via neutron-ovs plugin, I am getting maximum 3 Gbits throughput between vms on separate hosts (using set switch 2x 10gb NICs and ovs bond-slb on top). top, mtu 9k). That is quite disappointing. Is there any vxlan offloading mechanism possible? I've found information that it is planned on wiki of yours:

https://wiki.cloudbase.it/ovs-hyperv-architecture

I've tested disabling vmq and largesend offload as suggested in the same wiki. Not so much of a difference