Hi Xiaohu, On Sep 21, 2012, at 19:15, Xuxiaohu <[email protected]> wrote: ...
> Hi Kireeti, > > From a technical point of view, if the VXLAN and other similarities continue > to rely on multicast trees to emulate broadcast domains, rather than > resorting to a control plane protocol (e.g., BGP or ISIS) Well, the draft in question is trying to change that. > to realize VN membership auto-discovery , the globally significant VN ID > (e.g., VXLAN ID) on the data plane is absolutely necessary. Actually not. The management plane that programs the VNID could (in principle) use a different value per NVE. But practically, yes. > Otherwise (if they use a control plane protocol for VN membership > auto-discovery), the globally significant VN ID on the data plane seems not > much necessary anymore since the locally significant VN context ID (e.g., > MPLS label) can be signaled as well by using the same control plane protocol. Agreed. The benefit of each NVE doing the VNID allocation is that it can optimize for receipt and lookup. MPLS got a few things right, but this isn't always widely recognized. Kireeti > Best regards, > Xiaohu _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
