[ovs-dev] Bug#757761: openvswitch-switch: Please include ovsk-controller

[Tomasz Buchert]

On 11/08/14 14:28, Ben Pfaff wrote:
> On Mon, Aug 11, 2014 at 08:17:19PM +0200, Dariusz Dwornikowski wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA256
> > 
> > On 11.08.2014 17:38, Ben Pfaff wrote:
> > > On Mon, Aug 11, 2014 at 09:53:09AM +0200, Dariusz Dwornikowski
> > > wrote:
> > >> Source: openvswitch Severity: wishlist
> > >> 
> > >> Dear Maintainer,
> > >> 
> > >> Tomasz Buchert and I are working on introducing mininet, an SDN
> > >> emulator (http://mininet.org/) to Debian. Mininet heavily depends
> > >> on ovsk to provide OpenFlow switch but also an OpenFlow 
> > >> controller.
> > >> 
> > >> Ovsk starting from version 2.1 ships ovsk-controller as
> > >> test-controller and resides in tests/ directory of the main
> > >> source.
> > > Why can't mininet use a real controller instead of the useless
> > > test program from OVS?
> > 
> > Mininet is just an emulator, using test controller from OVS makes it
> > easy to start with and use by people who start with OpenFlow and SDNs.
> > 
> > Using an OVS controller is one of the major options in mininet, and in
> > fact the default behavior. You can always use external controller with
> > - --remote option in mininet.
> > 
> > If we ship mininet package without having OVS controller to use, we
> > will be shipping partially unusable software.
> 
> I don't understand why the OVS controller is useful.  OVS has a larger
> set of features, and performs better, with no controller, than it does
> with the test-controller.  Why do mininet users want to use
> ovs-controller instead of no controller at all?
> 

Hi Ben,
mininet users want to see mininet work - I doubt that they care, at least
at the beginning, how it works internally.

Mininet is basically an OpenFlow testing framework and therefore *always*
runs with an OpenFlow controller and hence *always* needs one. Therefore,
replacing it with "OVS without OF controller" is not feasible and probably
discouraged by the architecture of mininet.

Also, the "test-controller" was called a "reference controller implementation" [1]
at one point. Do you provide such a reference implementation right now?
If not, can you tell us how we can run a simple MAC-learning OpenFlow-controller
to manage multiple switches? Or OVS is not a right tool to do that?

Cheers,
Tomasz

[1] http://openvswitch.org/cgi-bin/ovsman.cgi?page=utilities%2Fovs-controller.8