On Sun, Feb 08, 2009 at 10:34:06PM +0100, Remko Nolten wrote: > Hi! > > For a shared virtual hosting project with some friends we need a > pretty > specialized network configuration. Because we have virtually no > time for > experimenting (no pun intended), and the hosting organization has no > experience with libvirt/KVM configurations they suggested to try > and ask > here. > > So, this is the situation: We have a server with one physical > connection. Each user gets one virtual machine with its own IP- > address > (we have our own ip-range). The hosting organization gives us one > "main" > address 'A' with netmask 255.255.255.0 (/24) and a default gateway > for > our host machine. Besides this we get a number of addresses B1, B2, > etc. > with netmask 255.255.255.255 (/32) and without a default gateway. The > switch just routes the traffic to our server for each ip-address we > own > (A and Bn). > > According to our hosting organization, the normal solution is to > configure a loopback interface for each of the ip-addresses on the > server so you can configure your software just to use one of the > addresses. But we are not sure how this will work out in our virtual > environment. Our intuition tells us we need to use the routing > network > forwarding mode, but when we bring the server to the data center we > don't want to have any surprises. :) Hi there. This actually sounds like a fairly standard configuration. What you want to do is set the single NIC on your server to bridge, following the instructions at http://wiki.libvirt.org/page/Networking#Bridged_networking_.28aka_.22shar... . Each VM can then have its own IP, which in your case sounds like it will be static. Hope this helps, --Hugh >

On Mon, Feb 09, 2009 at 04:51:34PM +0100, Remko Nolten wrote: > Thanks for your response. I'm afraid this will not work since the ip- > addresses for the VM's have a /32 netmask and no default gateway > which > is required when you create a interface using a bridge. > > This how it works in a (non-virtualized) system at our hosting > organization: > For example, my server has one main ip-address, 10.255.255.0 on eth0, > and our hosting organization gives us 3 extra ip-addresses (so our > range > is 10.255.255.0/30) which are routed to 10.255.255.0. On our > server, we > can add the ip-addresses using: > > ip addr add 10.255.255.1/32 dev lo > ip addr add 10.255.255.2/32 dev lo > ip addr add 10.255.255.3/32 dev lo This is not quite correct. The physical interface (eg. eth0 in the case of linux) has an IP address in a different network (say, for example, 192.168.0.1/24). The /32's are routed to this address. Since the machine has a default route (eg. to 192.168.0.254), this works fine. Cheers, Paul 'WEiRD' de Weerd > So, our server gets the routed packages and will recognize them as > one > of his own which enables us to add listeners to the ip-addresses like > Apache or a mail-server. > > In our case, we don't want the server to route the packages to the > loopback interface (so that services can listen to it), but we want > the > server to route the traffic to the virtual machine. The more I > think of > it, I suspect we need the "route" network option like this: > > <network> > <name>local</name> > <bridge name="virbr%d" /> > <forward mode="route" dev="eth0"/> > <!-- dev is the device which is directly connected to the network > and > has ip 10.255.255.0--> > <ip address="10.255.255.0" netmask="255.255.255.4"> > <dhcp> > <range start="10.255.255.1" end="10.255.255.3" /> > </dhcp> > </ip> > </network> > > Is this correct? >