Actually, LISP-DDT is the son of LISP-Tree that we proposed in a peer-review journal and that was the result of years of research.
When we proposed it, Olivier and I started the discussion around a hierarchical mapping system, we wanted to have something: 1. Scalable 2. Fault tolerant and troubleshootable 3. Securable 4. Insensitive to configuration error we saw that a hierarchical system "a-la" DNS was something that was really answering these points. Then, based on the hard evaluation work of our collaborators (Lorand, Florin, and Albert), we have been able to prove how good such a system could be and the gain w.r.t. the other solutions proposed at that time. I have to admit that at that time, I was convinced that using DNS could make it. However, after implementation and experimentations, we have seen that it was not possible to keep DNS as-is and that thus the DNS protocol would have to be extended. As changing DNS is not an easy task, and has using DNS protocol was suboptimal in term of performance (encoding, parsing…), we finally agreed with the fact that it was better to have a new protocol that mimics DNS without being DNS and having its limitations in our case. The authors of DDT are very aware of all this and DDT is a good tradeoff between our research expectations and their operational needs discovered deploying LISP in lisp4.net. For more details, check Loránd Jakab, Albert Cabellos-Aparicio, Florin Coras, Damien Saucez, and Olivier Bonaventure. 2010. LISP-TREE: a DNS hierarchy to support the lisp mapping system. IEEE J.Sel. A. Commun.28, 8 (October 2010), 1332-1343. DOI=10.1109/JSAC.2010.101011 http://dx.doi.org/10.1109/JSAC.2010.101011 Damien Saucez On 12 Nov 2012, at 19:27, [email protected] (Noel Chiappa) wrote: >> From: [email protected] > > Hi, thanks for those comments - most of them seem desirable, although I may > have some comments later. One thing that caught my eye quickly: > > >> 11.2.1 Why not use DNS: > >> imho DDT has been developed to blow up the LISP work. Right? > > This makes no sense, either organizationally, or technically. First, on the > non-technical front, the people who proposed DDT are the same people who did > LISP origially. Why would they want to blow up their own system? Second, I > cannot see how the selection of DDT is a crippling blow, technically. > > Replacement of the ALT with 'something better' was necessary; the reasons > aren't given in detail in the draft (which is intended for a future audience, > not today/yesterday, which is why the details were left out). It turned out > that what I saw as the worst problem with ALT (traffic hot-spots at the root) > turned out to not be a problem yet (although in the long run, it would have > been a major problem). Rather, the problem that was starting to appear with > the ALT (I gather, from a brief conversation) was simply the configuration > management - one had to configure all the tunnels, etc, etc. > >> The given explanations do not convince me. > > As to why DDT was preferred to a DNS-based system (the people at Cisco > actually prototyped one, to see how it would do, compared to DDT), I can > assure you that the reasons given in the document are a reasonable short > summary of an extended discussion that went on over a couple of days. > > Needless to say, everyone who participated in that discussion will have their > own views as to the strengths and weaknesses of each, and some people may > have a different take on the benefits of DDT (and indeed, there were some > people who actually preferred the DNS-based approach), but I tried to > summarize the reasoning as best I could in the short amount of space I felt I > could devote to the question. > > (Perhaps a longer, more detailed discussion of why DDT is better than a > DNS-based alternative would be in order in the DDT spec, but I don't think > it's proper in an introduction to the entire system.) > > I think in general the overall cost/benefit was felt to favour DDT - the cost > being the necessity to write code and specs, and the benefit being that we > didn't have to kludge anything, to adopt a system that was designed for a > different use; we could do a system that was exactly what was best for LISP. > > You may disagree with that, and agree with those who preferred the DNS-based > system. But that's your judgement - the judgement of me, and a number of > others, goes the other way. > > Noel > _______________________________________________ > lisp mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/lisp _______________________________________________ lisp mailing list [email protected] https://www.ietf.org/mailman/listinfo/lisp
