The ILNP Locator is, in effect, an IPv6 Prefix. The important question is in the rules by which it is generated; that is simple to implement but perhaps a little more difficult to describe.
In theory, you could have a 1:1/onto mapping between any 64 bit prefix within an edge network and a different 64 bit prefix at its ISP. Speaking in practical terms, doing so means some form of configuration or state maintenance that maintains the mapping. A stateless version of that might be found in http://tools.oetf.org/draft-mrw-behave-nat66; neglecting the matters of checksum calculation (which are unnecessary in an ILNP world), the mapping between an internal and an external prefix is literally to overwrite some number of bits in the prefix. If for example the internal network is using a /48 ULA and the external network is using a /56 prefix, one would overwrite 56 bits of the prefix in either direction, leaving the remaining 8 bits as they stand. There is the question of topology obfuscation (which I consider a red herring but I hear folks talk about); an option would be to literally overwrite 64 bits as though the internal network was a flat /64; that has issues in pro ving that the EID is in fact unique within the internal network, but could be made to work. So, yes, a locator is in some sense a subnet, but can also be aggregates of subnets in various ways. That becomes, if you will, a site. > Q1: What is the gain of ILNP over the current system in terms of > its effectiveness in reducing the IDR table size? The way I see it, the fundamental benefit of GSE or any of its successors, of which ILNP is one, is that the edge network is able to operate as if its address was provider-independent, and the transit domain is able to operate as if the address is provider-allocated. Neither adds additional complexity, as compared to (say) shim6, which forces the edge network to accept the additional complexity of routing multiple prefixes for the same subnet, or a truly PI network, which forces the transit domain to enumerate edge networks. I think you will find, from the potaroo numbers, that there are on the order of 5000 transit networks in the world. Apart from BGP Traffic Engineering, which none of these address, that says that you should expect about 5000 PA prefixes, as opposed to O(10^6) to O(10^7) PI prefixes. There are of course some edge networks that really do work best with their own prefix at the edge; if you wave your hands and accept that anyone that has an AS number now falls in that category (which I would dispute, but that's another discussion), we're looking at O(30,000) such prefixes today. Again, the comparison is O(10^4) as opposed to O(10^6) or O(10^7). On May 26, 2010, at 4:47 AM, Dae Young KIM wrote: > Hi, > > Now that the decision is made for ILNP and there's not yet the ILNP WG > formed, would it yet be appropriate to raise some questions in regard > to ILNP to facilitate better and clearer understanding of the > technology? > > Although some previous responses have been sometimes something like > 'you read the draft', to some slow people like me, there still remain > curiosities not to be cleared only out of the given text. Or the text > is too long to know where to find the answer. > > As long as this is a research group(RG), I'd assume there'd be some > room for pedagogical purposes, I'd hope. > > So, could anyone bother to give some comments to my following line of > thoughts, please? > > o I understand the granularity of the Locator is a subnet. > o As seen from outside of a given site, the effective granularity > of the Locators stemming out is a site, for all such Locators will > eventually be aggregated to a number representing the very site in a > single entry in the IDR router table. > o Even with the current Internet, the whole network prefixes > belonging to a well-behaved site would be aggregated to a single > shorter prefix representing the very site in a single entry in the IDR > router table. (.. except for some multi-homed hosts inside the site.) > > Then, here's the question: > > Q1: What is the gain of ILNP over the current system in terms of > its effectiveness in reducing the IDR table size? > > Perhaps, so obvious for the author or pros, but not to me at the moment, yet. > > -- > DY > _______________________________________________ > rrg mailing list > [email protected] > http://www.irtf.org/mailman/listinfo/rrg http://www.ipinc.net/IPv4.GIF _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
