I don’k know really why, out of the 2 servers I rent, one works like a charm (the slower one of course) the other one has terrible throughput when I download a file from the server to workstations in my LAN. After installing iperf and running various tests, I deducted the slow downs affected only VMs. When I was testing flows between physical hosts I had the correct results. Even when the communication is tunnelled with GRE over IPSEC. For the fun, I wanted to try if the problem was the routed network on Libvirt, as I know it’s not a common setup maybe it was the cause of my issues.

Following are the instructions I used to replace the routed network with one managed by OpenvSwitch.

Installing OpenvSwitch

  • Install the packages
[root@db-xc1 696 ~]# apt install openvswitch-switch

Define the new network using Libvirt

  • Delete the old network
[root@db-xc1 696 ~]# virsh net-destroy VMNetwork
  • Create the xml entry for the new network
<network>
  <name>OVSNetwork</name>
  <forward mode='route'/>
  <bridge name='br10' stp='on' delay='0'/>
  <ip address='172.16.4.1' netmask='255.255.255.0'>
    <dhcp>
      <range start='172.16.4.100' end='172.16.4.199'/>
    </dhcp>
  </ip>
  <virtualport type='openvswitch'/>
</network>
  • Define the new network and start it automatically at boot
[root@db-xc1 696 ~]# virsh net-define /root/ovsnetwork.xml
[root@db-xc1 696 ~]# virsh net-autostart OVSNetwork
[root@db-xc1 696 ~]# virsh net-start OVSNetwork

Update the Bird configuration file to adapt the routing

  • Update on the /etc/bird/bird.conf file :
protocol device {
        scan time 60;
        primary 172.16.4.0/24;
}

protocol direct {
        interface "br10";
}
  • Restart the bird daemon and check the routing table
[root@db-xc1 35 ~]# service bird restart
[root@db-xc1 35 ~]# birdc show route

Update the VMs

After enabling the new LAN, don’t forget to change all VM to the new network. Once done, you can check again the VM are accessible and the bandwidth is corresponding to the real values :)