On Thu, Jul 17, 2014 at 10:39 PM, Dino Farinacci <[email protected]> wrote:
Say you have a cluster of VMs that work together at the applicaiton-layer. > And you want to move them together to a cloud provider. In this case the > EIDs of the VMs in the cluster are part of the same power-of-2 prefix. If only the EIDs already had been assigned out of the same power-of-2 prefix since the cluster of VMs is part of the usual Internet. What if they had not? > In this case, that is an EID-prefix that moves so the new set of RLOCs are > the one that reside at the cloud provider. > An EID-prefix needs only one RLOC, doesn't it? The whole EIDs in the same site would be mapped to the one single RLOC (in the basic model) for the purpose of inter-site routing... > Another related question might be.... is there any compelling reason why > EIDs within a site should better be assigned in the same power-of-2 prefix? > > If they fate-share RLOC changes, they should be assigned in power-of-2 > blocks. > Why should EIDs fate-share RLOC changes? EID are for e2e connection while RLOCs are for routing... > If the EID would be assigned to a node, it wouldn't matter much. Or is it > to save entry space in the mapping server? > > The only real issue with assigning addresses to nodes versus interfaces > (this is a general comment), is if the routers know directly how to reach > the address. When the address is assigned to an interface, routers are > directly attached to the subnet of node's interface address. When it is > assigned say to a loopback interface, the routers need more information to > know what physical interface the node is reachable on. > Not a problem in link-state routing like ISIS or OSPF... or..? -- DY
_______________________________________________ lisp mailing list [email protected] https://www.ietf.org/mailman/listinfo/lisp
