Open Stack

On Mar 18, 2014, at 7:40 AM, Roman Verchikov <rverchikov at mirantis.com<mailto:rverchikov at mirantis.com>> wrote:

Hi stakers,

We’re trying to replace dnsmasq-supplied DHCP for tenant VMs with a vendor’s baremetal DHCP server. In order to pass DHCP request to a vendor’s server and send DHCP response back to VM we decided to add another OVS bridge (we called it br-dhcp), connected to integration bridge (br-int), which will have OVS rules connecting VM’s MAC address with br-dhcp port. In this scenario DHCP response will only find it’s way back to a VM if VM has globally-unique MAC address.

My questions are:

  *   is having code which generates globally-unique MACs for VMs acceptable by the community at all?

This question tends to pop up from time to time and there are valid deployment and usage scenarios where you would want to assign the same MAC to multiple ports.


  *   is there a better solution to the problem (we also tried using dnsmasq as a DHCP relay there)?

That answer really depends on a number of factors.
 - Are the IP allocations being handled inside or outside of Neutron?
 - Do you allow different networks to have overlapping IP ranges?
 -

If it is outside of the OpenStack deployment then your code can use flow mods with you br-dhcp. If Neutron is managing the allocations or you allow overlapping IPs, you probably want to consider implementing a driver for the DHCP server.

mark

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140318/023711b4/attachment.html>