A centralized architecture for a single domain is the way things will be in practice (already are). One way or the other, non-virtualized network devices (physical switches etc) will have to fit into that model.
Now we can debate the definition of a domain is :) It may be defined by the scope of a centralized control plane? Somesh From: [email protected] [mailto:[email protected]] On Behalf Of Lucy yong Sent: Tuesday, July 23, 2013 8:15 AM To: Aldrin Isaac; Kireeti Kompella Cc: [email protected]; L3VPN; Yakov Rekhter; [email protected]; Xuxiaohu; Luyuan Fang (lufang); UTTARO, JAMES Subject: Re: [nvo3] The possibility of using global MPLS labels as VNIs ... for l3vpn For distributed based architecture, the local context ID has a lot of benefits. For centralized (or SDN) architecture, will the global context ID have advance? Another way to ask this is: is distributed architecture better than centralized architecture for VPN or virtual network overlay? Lucy From: [email protected]<mailto:[email protected]> [mailto:[email protected]] On Behalf Of Aldrin Isaac Sent: Tuesday, July 23, 2013 8:00 AM To: Kireeti Kompella Cc: Luyuan Fang (lufang); L3VPN; Yakov Rekhter; [email protected]<mailto:[email protected]>; Xuxiaohu; [email protected]<mailto:[email protected]>; UTTARO, JAMES Subject: Re: [nvo3] The possibility of using global MPLS labels as VNIs ... for l3vpn +1. Network operators lose flexibility with Global VNIDs. Virtual topology will be fine tuned on top of VNIDs by squeezing a square peg (ACLs) into a round hole (where it's otherwise not needed). I think local context IDs will make a come back once we regain our sanity. :). On Tuesday, July 23, 2013, Kireeti Kompella wrote: Hi Xuxiaohu, Sorry for the previous empty email. On Jul 23, 2013, at 4:51, Xuxiaohu <[email protected]<javascript:;>> wrote: > The reason that I started this discussion is to make sure whether the Virtual > Network Context Identification contained in the data packet is REALLY > required to be globally unique in some cases. No. There's nothing useful that I know that one can do with global IDs in the data plane that one cannot do with local IDs, and local IDs are easier to allocate and manage. So, I'd be going the other way and advocate locally significant VNIDs for VXLAN and NVGRE, not global IDs for VPNs. (And I do) Kireeti _______________________________________________ nvo3 mailing list [email protected]<javascript:;> https://www.ietf.org/mailman/listinfo/nvo3
