Re: [nvo3] Draft NVO3 WG Charter

[Robert Raszuk <[e-mail]>]

Hi Yakov,

> What are the specific *technical* reason(s) why MPLS over GRE is
> a "non-starter" for (a) ToR switches, (b) datacenter access switches,
> and (c) hypervisor softswitches ?

I don't think this is right question to ask. I personally see no 
technical reasons why MPLS over GRE could not be used in (a) or (b) or 
(c). In fact as one could notice from my former posts I think this is a 
very good idea to use it.

However the right question here is the choice of signalling and control 
plane such that either (a) or (b) or (c) knows how to build GRE header 
or what value of MPLS label to impose. I think this is the crux of focus 
of this potential WG.

IMHO the better questions to ask this WG are ...

* If we choose draft-marques-l3vpn-end-system to be the signalling 
protocol for nvo3 would we get general agreement that all (a) ToR 
switches, (b) datacenter access switches and (c) hypervisor 
soft-switches would support it ?

* Do we need one signalling protocol for all (a) or (b) or (c) or would 
we be fine with more then one depending on the level of virtualization 
boundary

* Do we in general agree that p2mp nature of BGP distribution (even with 
RTC) which is in core of draft-marques-l3vpn-end-system proposal is a 
good fit for DC L2 and L3 virtualization requirements ?

* For short lived VM based virtualization requirements (job processing 
application) especially across DCs are we really convinced that dynamics 
of today's BGP (where per prefix priority is simply not implemented 
anywhere and where fair treatment to all prefixes is preferred) is the 
best mechanism we could come up today to this problem space ?

Best regards,
R.