On Tue, Nov 11, 2014 at 10:52 PM, FengYu LeiDian <fengyuleidian0...@gmail.com> wrote: > 于 2014年11月06日 23:27, Jesse Gross 写道: > >> On Wed, Nov 5, 2014 at 10:58 PM, FengYu LeiDian >> <fengyuleidian0...@gmail.com> wrote: >>> >>> 于 2014年11月06日 00:08, Jesse Gross 写道: >>> >>>> On Tue, Nov 4, 2014 at 11:03 PM, FengYu LeiDian >>>> <fengyuleidian0...@gmail.com> wrote: >>>>> >>>>> >>>>> Hi >>>>> >>>>> Env: redhat 6.4, OpenvSwitch-2.1.2, using native >>>>> datapath/linux/openvswitch.ko module >>>>> >>>>> VM1 on host1, VM2 on host2, host1 and host2 are connected by a switch. >>>>> both VM enable virtio/vhost when launching VM >>>>> >>>>> case1: >>>>> VM -> tap -> ovs-bridge -> eth1 >>>>> >>>>> case2: >>>>> VM -> tap -> ovs-bridge -> vxlan -> eth1 >>>>> >>>>> When using vxlan in case2, iperf performance drop 60%, >>>> >>>> >>>> >>>> Your NIC probably doesn't support offloads (checksum, TSO, etc.) in >>>> the presence of VXLAN. >>> >>> >>> >>> This is the default features supported by my NIC 82599 both in case1 and >>> case2. >> >> >> I can assure you that this NIC does not support VXLAN and the stack is >> being forced to do segmentation in software. > > Sorry for the late reply. > > Do you mean NIC supports VXLAN offload? > Yes, my 82599EB does not have this feature. > > > After some debugging, when VM sending small packet, > > In case2: performance 80Mbps > VM -> tap -> ovs-bridge -> vxlan -> eth1 > > tcpdump tap0 observed packet size bigger than 1000 Bytes because > of VM eth TSO involved. > > While tcpdump in eth1 observed small packets with the same size 192Bytes. > Looks like skb segmentation happens somewhere before vxlan. > > I tried to set vxlan1 interface MTU by below, no luck. > ovs-vsctl add-port ovs-br0 vxlan1 -- set interface vxlan1 type=vxlan > options:remote_ip=192.168.1.5 options:key=99 options:dst_port=4789 > options:df_default=true mtu=9000 > > > In case1: performance 400Mbps > VM -> tap -> ovs-bridge -> eth1 > > Configuration is same as Case2, only without vxlan interface. > Both tap and eth1 observed bigger packet size.
The packets appear to be larger because segmentation is being done on the NIC instead of in software. _______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
