[ovs-discuss] UDP loss rate in OVS 2.5

Amru OM sdn.innovation at gmail.com
Sat Jul 2 14:54:27 UTC 2016


Greetings,



I have installed the latest OVS 2.5 in each physical server of my testbed.
The physical servers running CentOS Linux release 7.2.1511 as the host
operating system. Also, each host has one VM running via KVM.  The VM in
each server is connected to OVS through vNIC (e.g., tap0). Here is the
command line that I used to configure and start-up the VMs



*qemu-kvm -m 2048 -netdev
tap,id=hostnet0,script=/etc/ovs-ifup,downscript=/etc/ovs-ifdown,vhost=on
-device virtio-net-pci,
netdev=hostnet0,id=net0,mac=52:54:00:af:87:57,bus=pci.0,addr=0x3 -drive
file=/var/lib/libvirt/images/fedora.img -cpu host -smp cores=1,threads=2*



the OVS (in each server) and the host NIC are configured as follows:



*ovs-vsctl add-br br0*

*ovs-vsctl add-port br0 eth0*

*ifconfig eth0 0 up*

*ifconfig br0 10.0.0.x netmask 255.0.0.0 up*



eth0 above-mentioned is the physical NIC in the host and its link-speed is
56 Gbits/sec.

 After configuring and running OVS and VM in each host, I tested the
bandwidth between two VMs located in different hosts via *Iperf*. When I
test the UDP bandwidth, where packets have large payload ( >= 10KB), the
test results show high packet loss rate.  For example, when I run *Iperf*
as follows:



*Iperf  -u -c 10.0.0.x  -l 10K  -b 56G*



The test result of UDP bandwidth is 2.88 Gbits/sec, and packet loss rate ~
72%. This poor performance only occurs with UDP, but not with TCP. I just
wonder if I’m missing something here.





Please advise and thank you in advance.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160702/070abec0/attachment-0002.html>


More information about the discuss mailing list