[ovs-discuss] Rhel 6.1 - openvswitch 1.2.2 - network commutation between physical and virtual - doesn't work

Leland Vandervort leland at DEV.DISCPRO.ORG
Fri Sep 30 12:00:20 UTC 2011


Benoit, 

I presume that the physical interface in question is connected to a physical
switch with dot1q tagging in operation on the physical port?  If so, is it a
Cisco switch by chance?  And if yes, is the port configured to use vlan2 as
the native vlan?  We had a similar issue with the bridge in Xen where the
Xen Bridge tagged the ARP request when sending it to the physical switch,
but since the VLAN in question was the native vlan for the trunk, the
response was not tagged.  Xen Bridge had no idea what to do with the packet
because it wasn¹t tagged, and there was no support in the Xen bridge for a
³native² vlan definition.

As a work-around, we set the physical switchport to tag ALL vlans by setting
the native vlan to 1 (which is not being used anyway).

The other potential reason could be that although the switchport is
configured as a trunk, it also has an access vlan configuration ... or
finally, simply that the physical switch port isn¹t configured as a dot1q
trunk, so doesn¹t tag any frames.

Things to check and confirm.

Regards, 

Leland



Le 30/09/2011 13:49, « Benoit ML » <ben42ml at gmail.com> a écrit :

> Hi,
> 
> Some more information :
> When I use TCPDUMp on the physcal interface user by openvswitch :
> - I see the vlan ID  when the packet leave openvswitch (ARP request)
> - I didn't see the vlan ID when the packet arrive (ARP respond)
> 
> It's like the  vlan was not transmit to openvswitch  (I clearly see that the
> ARP answer is dropped by ovs)
> 
> Regard,
> 
> 2011/9/30 Benoit ML <ben42ml at gmail.com>
>> Hello,
>> 
>> We suffering a problème about network switching between a physical world and
>> a virtual world :
>> 
>> We have a x86 server, with 10Gb cards.  On the vswitch we have  add some
>> cards to do the switching :
>> ovs-vsctl  add-bond brCentral ovsbond0 eth4 eth5 bond_mode=active-backup
>> bond_updelay=10 bond_downdelay=10 lacp=off trunk=[0,2,3702]
>> 
>> But appears a broblème :  the switching works very good for untagged frames
>> but doesn't work fort tagged vlan 2 frames ... So the VM never see the
>> response for a simple arp request :(
>> 
>> 
>> I've see the "ovs-vlan-bug-workaround eth4 on"  but :
>> ovs-vlan-bug-workaround: operation failed (kernel does not support the VLAN
>> bug workaround)
>> 
>> We have tested with "Emeluex oneConnect 10Gbps", module be2net and with
>> braodcom bnx2x ...
>> 
>> 
>> Any idea plz ?
>> 
>> 
>> Software version
>> RHEL6.1
>> kmod-openvswitch-2.6.32-131.12.1.el6-1.2.2-1.el6.x86_64
>> openvswitch-1.2.2-1.el6.x86_64
>> kernel-2.6.32-131.12.1.el6.x86_64
>> modinfo be2net
>> filename:       
>> /lib/modules/2.6.32-131.12.1.el6.x86_64/kernel/drivers/net/benet/be2net.ko
>> license:        GPL
>> author:         ServerEngines Corporation
>> description:    ServerEngines BladeEngine 10Gbps NIC Driver 2.103.298r
>> version:        2.103.298r
>> srcversion:     538F174333FC55CF8B6A4BD
>> alias:          pci:v000010DFd0000E220sv*sd*bc*sc*i*
>> alias:          pci:v000019A2d00000710sv*sd*bc*sc*i*
>> alias:          pci:v000019A2d00000700sv*sd*bc*sc*i*
>> alias:          pci:v000019A2d00000221sv*sd*bc*sc*i*
>> alias:          pci:v000019A2d00000211sv*sd*bc*sc*i*
>> depends:        
>> vermagic:       2.6.32-131.12.1.el6.x86_64 SMP mod_unload modversions
>> parm:           rx_frag_size:Size of a fragment that holds rcvd data.
>> (ushort)
>> parm:           num_vfs:Number of PCI VFs to initialize (uint)
>> parm:           multi_rxq:Multi Rx Queue support. Enabled by default (bool)
>> 
>> 
>> 
>> 
>> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20110930/066d0729/attachment.html>


More information about the discuss mailing list