the reason folks came up with VxLAN is to solve an urgent
problem (VM placement) while leverage existing data center
L2/L3 switches.
1. We are having a hard enough time agreeing on naming while
IAAS providers have to compete with likes of Amazon
2. What other existing encapsulation would do the work with
legacy equipment?
a. MPLS is not implemented in DC switches
b. Implementations of GRE (in merchant ASICs) do not support
a tenant field
c. How would ECMP be supported in existing equipment
d. And we could go on to other encaps (not widely implemented,
L2 only etc)
3. Not require a new control plane implementation in the virtual
switches - which we seem to be spending a lot of effort on
-- When the entire software driven data center is orchestrated
centrally, why would hypervisor/vswitch vendors all of a
sudden not use all of the knowledge they already have and
implement and depend on yet another mechanism (even if
as cool as BGP).
Unless we start looking at the problem from the point
of view of centrally managed/orchestrated software defined data center
instead of distributed protocol problem, we might
end up with something no-one really cares about.
Somesh
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of
> Yakov Rekhter
> Sent: Friday, September 21, 2012 12:04 PM
> To: Kireeti Kompella
> Cc: [email protected]
> Subject: Re: [nvo3] unified encapsulation headers for both Layer2 and
> Layer3 VPN overlays?
>
> Kireeti,
>
> > > Kireeti,
> > >
> > >> Hi,
> > >>
> > >> For a number of reasons, including the current "vigorous
> discussion" on
> > >> multiple encaps for L2-in-L3, I would STRONGLY URGE the group to
> not go
> > >> there.
> > >>
> > >> There are more than enough L3-in-L3 encaps already: sufficient
> unto
> > >> the day
> > >
> > > I certainly agree with you that there are more than enough L3-in-L3
> > > encaps that are IETF standard.
> > >
> > > But do you think we have a shortage of L2-in-L3 encaps that are
> > > IETF standards ?
> >
> > Not at all! Sadly, there seems to be more interest in coming up
> > with new encaps than in coming up with a control plane that can
> > actually deal with the problems of large scale multi-tenant data
> > centers. However, taking these new encaps as a pragmatic reality,
> > having a single control plane that can signal across all of them
> > is a Good Thing.
> >
> > Just to be clear, these new encaps are not yet IETF standards (but
> > it seem s likely that some of them will be.)
>
> I would like the WG chairs (and the appropriate ADs) to comment
> on what criteria will be used to decide which these encaps be
> IETF standards.
>
> Yakov.
> _______________________________________________
> nvo3 mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/nvo3
_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3
