ProCurve 1800-8G, stripping off VLAN tags from the inbound Ethernet frames

[vap]

Hi,

In our CI environment we have tens of testbeds running different configurations. Each testbed is normally connected to a test PC but our purpose was to consolidate all these test PCs by mapping one KVM virtual machine for each testbed. The requirements say,  that the testbeds need to to be able to connect to the VMs using the original static IP addressing scheme( making the testsbeds VLAN aware is not desirable). All testsbeds have the same IP address

Using a VLAN per VM-testbed pair I was able to get the system up and running IF the  testbed was working as an endpoint for the VLAN. On the server-side each VM is connected to their own virtual switch(Linux bridge) and each of these are on a different VLAN. So basically all VMs are unaware about the underlying VLANS because the virtual switch works as an VLAN endpoint.

I would want that it would be possible to strip off the VLAN tags from all inbound ethernet frames accessing a port connected to a testbed. This way the frame would reach the testbed untagged. The testbed sends only untagged frames and converting those to VLAN tagged frames isn't a problen. Stripping the tags from inbound frames is the problem because I don't seem to find this kind of option from the ProCurve 1800-8G.

Any suggestions about the solution or possibly some other switch which would support this kin of feature?

Thanks in advance.

[paulgear]

Hi vap,

As a general rule, please ask questions about 1800 series systems in the V-series forum: http://h30499.www3.hp.com/t5/Web-and-Unmanaged/bd-p/switching-v-series-forum

With regards to your specific question, you can get the KVM host to handle everything. Set each VLAN as tagged on the host port, and attach your VM bridge to the VLAN interface. Here's how i do it with Debian's /etc/network/interfaces:

...
auto vlan3
iface vlan3 inet manual
vlan-raw-device eth0

auto br3
iface br3 inet manual
bridge_ports vlan3
bridge_stp off
bridge_maxwait 1
...

In this example, eth0 is the port on the switch which is tagged for VLAN 3. Any VM attached to br3 will be untagged on VLAN 3. You can add this for as many VLANs as you need, and the host will take care of the tagging.

[vap]

Hi paulgear,

Sorry for describing the problem unclearly. The VM side is fine and has been defined in the similar manner as your setup. The testbed itself is the problem because it doesn't understand the VLAN tagged frames received from a different port. The port  which is attached to the hypervisor(and VMs through virtual switches).

And sorry for posting on the wrong forum.

Br,

vap

[paulgear]

Hi vap,

I still don't understand - what is your "testbed"?  Whatever it is, if you want to make it non-VLAN-aware, simply make it untagged on, say, VLAN 3 and then set the trunk port on the VM host to have VLAN 3 tagged.  The switch will take care of making sure they are part of the same broadcast domain.

[paulgear]

Sorry, i've just re-read your first message, and it seems to indicate that the "testbed" is a VM on the host, not another network node.

But the fact remains, that with the setup we've described on the VM server, all you need to do is tag all of the correct VLANs on the server trunk from the switch side, and those VMs will be in the correct VLANs.