Sorry - a couple more comments in-line: On Wed, Jun 22, 2011 at 7:02 PM, Dino Farinacci <[email protected]> wrote: >> From the discussion on draft comments, I have the following basic >> question: >> >> Is a value A is assigned to either the EID space or the RLOC space? >> Could site X have an EID with value A while site Y (or >> even a non-LISP) has an RLOC (or globally routable address) with the >> same value A? > > Architecturally, yes, the value A can be an EID and an RLOC. In practice, > no, for IPv4 and maybe for IPv6. Let me explain. > > Since there are two namespaces for each of IPv4 and IPv6, it means, for the > case of IPv4, there are two 2^^32 number assignment spaces. But we don't > have two allocation authorities, one for each, so the addresses will be > assigned from one 2^^32 pool and be used as either an EID or an RLOC > depending if the site has converted to being a LISP site. > > For IPv6, if we had a PI allocation authority, then it would hand out EID > prefixes to end sites. If we also had a PA allocation authority, then it > would hand out RLOC addresses to infrastructure providers. In this case, if > the two authorities acted independently, then the same value could be > assigned for each namespace. > > This is not a problem to duplicate the address in each namespace. But I do > believe for operational sanity it would be nice to look at logs, debugs, or > whatever, see an address and decipher it is an EID versus an RLOC. This is > one of the reasons the working group wants to request an IANA assigned /12 > or /16 (not decided yet I think).
I believe there are assumptions in the drafts now that prevent duplicating an address in each namespace. >> For instance, consider deploying an IPv4 LISP site now. Could one >> take an IPv4 prefix already used >> globally by a different company/site - and use it for my new LISP site >> as an EID prefix? > > No because there is one allocation authority and it is enforcing a unique > address allocation policy. If you decided that EIDs were a real separate namespace - managed by another entity? >> Do all the drafts always check for the IP address in the mapping >> database to see if it is an EID? I recall seeing some >> cases of checking the global routing table - but that could be bad >> memory at this point. > > If you look in the ALT routing table and find a prefix, it is an EID. That > is an example of looking in *a routing table*. But that is part of the > mapping database system. So it is one in the same. I believe it was, and certainly my interpretation, was to look it up in the "Global Routing Table". For consistency, if a lookup is done in the Routing Database, then the same terminology should be used. >> Could a host in a LISP site send to an IP address as an EID and the >> same IP address as a globally addressable (or routable)? > > A host sends to destinations. So it doesn't know one from the other (a > feature). So yes, both a non-LISP site host and a LISP site host can talk to > both a non-LISP site and LISP site destination. Let me provide examples, since I strongly think the answer is NO and feel you have side-stepped the question into vague generalities that ignores the issue. An end-host in a LISP site puts a destination of A in its packet. The ITR checks for A in the Mapping Database & finds it. Therefore, the ITR encapsulates the packet and sends it to an RLOC specified in the locator-set. There is NO WAY for a LISP end-host to send a packet to the globally addressable destination A - unless the EID A and the globally addressable destination A refer to the exact same host. Similarly, an end-host that is not in a LISP site puts a destination of A in its packet. If an PITR tries to claim value A because it is an EID, then depending on route-preferences, the non-LISP end-host may be able to send to globally addressable destination A or its traffic might end up going to the advertising PITR. >> I am confused because "architecturally" I believe the EID space and >> the RLOC space are separate namespaces - but in practice >> in the drafts, it seems that a given value must belong to a single >> entity, whether it is used as an EID, globally addressable, or both. > > That is what you get when you build an architecture after the network is > built. ;-) It is an amusing challenge - but basic architectural assumptions need to be clearly documented as the reality they are. Alia > Dino > >> Is this clearly specified anywhere? What am I missing? >> >> Alia >> _______________________________________________ >> lisp mailing list >> [email protected] >> https://www.ietf.org/mailman/listinfo/lisp > > _______________________________________________ lisp mailing list [email protected] https://www.ietf.org/mailman/listinfo/lisp
