On Thu, Mar 22, 2012 at 3:57 AM, Lior Gradstein <[email protected]> wrote:
> Hello,
>
> I'm a total newbie on Open vSwitch, so my question may be stupid, but I
> couldn't find the answer in the documentation. As I read, Open vSwitch
> supports bonding of virtual interfaces. Is it possible to use it to bond
> several interfaces (i.e. two interfaces, which may be virtual, like GRE or
> PPP) that go through Internet (to another Open vSwitch I guess), with each
> interface having a different output gateway address?
> I am already doing it with the native bonding module from the Linux kernel.

In theory it's possible to run bonding over OVS GRE ports, in practice
it's not usually very interesting unless you only care about load
balancing or are willing to spend some time writing flows because GRE
ports don't have a carrier status.

> On one side, I have two PPPoE links (through one NIC) each going to a
> different PPPoE server, each PPPoE being encapsulated in a separate GRE
> tunnel (the reason for GRE is because on the receiving side, there is just
> one NIC, and to be able to use the Linux bonding I need two links on both
> sides). So I'm ending up bonding two GRE interfaces on each side :-) From
> this gateway it goes through Internet to another machine, where I have one
> NIC with the two corresponding GRE interfaces, bonded with the Linux kernel
> bonding module.

I can't follow this description but it sounds overly complex.

> Is it possible to replace the bonding with Open vSwitch to achieve something
> cleaner? Also, does Open vSwitch balance each packet (like the bonding
> module does), or only a session (by session I mean one flow until the RST or
> FIN flag) through each interface?

Open vSwitch uses flow based balancing, per-packet balancing is likely
to result in reordering problems.  It's not clear to me which part of
this you think that OVS will simply over Linux bonding.
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to