[ovs-dev] [PATCH] vswitch: Consistently set Nicira OUI.

Ben Pfaff blp at nicira.com
Sat Feb 6 03:58:38 UTC 2010


On Fri, Feb 05, 2010 at 08:30:10PM -0500, Jesse Gross wrote:
> On Tue, Feb 2, 2010 at 10:27 AM, Jesse Gross <jesse at nicira.com> wrote:
> 
> > On Mon, Feb 1, 2010 at 8:50 PM, Justin Pettit <jpettit at nicira.com> wrote:
> >
> >> On Feb 1, 2010, at 4:43 PM, Jesse Gross wrote:
> >>
> >> > The bottom line is that we should be consistent everywhere because
> >> otherwise we end up with the worst of both worlds.  Personally I prefer have
> >> the OUI set since it is nicer when looking at packet traces.  It's also
> >> friendlier to the rest of the world.  VMware sets their OUI on addresses
> >> they generate.
> >>
> >> Is there a standard for randomly-generated "official" addresses?  We set
> >> that top bit of the third byte to indicate that the address is randomly
> >> generated, so that we could also assign real, static addresses if we need to
> >> down the line.  The IEEE doesn't like handing out additional OUIs unless you
> >> can prove that you've actually run out of the first block you were given.
> >>  I'm just curious if there's a better approach than my
> >> off-the-top-of-the-head suggestion of setting a single bit to indicate this.
> >>  I'd think VMware or someone has given this a little more thought.
> >
> >
> >  I don't believe that there is any standardized way of doing this.  VMware
> > has been allocated four OUIs, which they use in different products.
> >  However, even within ESX they use two different OUIs: one for generated
> > addresses and one for manually configured addresses.  The generated
> > addresses are not random though: they are based on the IP address of the
> > machine, a hash of the config file, etc.
> >
> > I think we could make a pretty good case getting another OUI if necessary
> > in the future, so I don't think that is a big concern.  Setting the top bit
> > is fine with me too (we are currently setting the top two bits in one place
> > but none in another).  Also, these addresses are only used for traffic
> > actually generated by an OVS interface, not every VM (this is different from
> > VMware), so the chance of a collision isn't too high.
> >
> 
> Can we make a decision on this?  Like I said before, what we have now is
> pretty much the worst of all worlds.

I didn't realize that you were waiting for a decision.

Speaking for myself, your last round of arguments was convincing enough
for me, so I'm comfortable going ahead with always using the Nicira OUI.
I'd like it if we set at least one bit to a constant value, maybe two.




More information about the dev mailing list