Divorcing Registries from TLDs – fostering true competition and innovation.
Professor Dennis Carlton recites the economist’s tenet that competition enhances consumer welfare. But he gives too little weight to the converse: that monopoly harms welfare. A top-level domain (TLD) is a monopoly over a swathe of meaning: all of world-wide commercial activity (.com); everything to do with automobiles (.cars); all of the Catalan language and culture (.cat); the whole of Germany (.de). Any registry that owns such a territory then simply extracts rent (through the agency of registrars) for the use of subsidiary monopolies: the second-level domains names (domains, for short). To increase competition and innovation – and hence consumer welfare and economic productivity – would it not be more effective to break up these monopolies rather than to create extra ones? There is no technical reason why all domains in a given TLD must be let by a single registry. Just as multiple registrars compete for domain registrations within the same TLD, so multiple registries should compete, within the same TLD, for the business of registrars. Of course, there are powerful institutional and contractual barriers to be broken down. Registries will wail and moan and fight to the death as their monopoly is threatened. And, yes, a little bit of new software will need to be written and tested. But conceptually and technically there is nothing hard about it. Second level domain names are monopolies within a TLD, because the domain name system (DNS) requires the combination of domain.TLD to be unique. In the real world, names are not unique: - Firstname.Surname rarely identifies a unique individual - Registered trademarks are unique only within the trades for which, and the jurisdictions in which, the marks are registered. Thus, in the UK, Amazon the online retailer can coexist with Amazon the maker of bilge pumps and Amazon the maker of laundry equipment. But there is only one amazon.co.uk - Company names need not be trademarks. Just being registered as a company name requires a degree of uniqueness within a given jurisdiction. This tends to give rise to convoluted names, like Amazon Windows (Hull) Ltd, as distinct from Amazon Tree Houses Ltd or, indeed, Amazon Ltd. But across jurisdictions (generally countries, but individual states in the US) the same name may indicate different companies. Yet each owner of these names has a legitimate claim to the matching domain name. Mapping the mess of real-world, non-unique names to the unique identifiers required by the DNS, results in more losers than winners – for there can be only one winner in any one TLD. This fact is the root cause of the demand for more TLDs. However, the magnitude of real-world name duplication, and the richness of the real-world mechanisms that have evolved to deal with it, cannot be accommodated simply by increasing the number of TLDs. A real-world name is unique only in a context. The context may be modelled as multiple, intersecting, classification trees. For example, in trademarks – a reasonably well-defined name space – a name is unique at a particular intersection of jurisdiction x trade. A single-rooted tree, like the DNS, cannot hope to begin to deal with this. So let’s deal with it somewhere else. There is no problem in computing that cannot be solved by adding an extra level of indirection. The DNS already provides a level of indirection, by mapping domains to IP numbers. This allows references, such as bookmarks, hyperlinks, email addresses to stay constant, while the IP address, where the services or resources can be found, changes. We don’t want to lose that. But we can lose the historic feature whereby references are meaningful (domain) names. Meaningful names were a vital feature in the early days of the DNS, when there were no search engines and clients were text-only, resource constrained consoles. Those constraints have gone and we can build any required data storage and intelligence into the client, into servers, and, indeed, into points in between. So let’s abstract meaning from the domain ‘name’, leaving just a unique but meaningless identifier that the DNS then maps to the IP address. Let’s relocate the meaning to a new layer where real-world names are allowed to exist in all their duplicate glory and mess. A new level of indirection can then map a selected name to a unique identifier in the DNS, from where established DNS functionality maps to the IP address. This will alleviate the scarcity of names by allowing real-world levels of non-uniqueness. String contention and conflict resolution no longer have to be imported into ICANN procedures, they can stay where they are: in real world company and trademark registration and lawyering. DNS as is Proposed +------------+ +------------+ | domain.TLD | | real name | +------------+ +------------+ | | v v +------------+ +------------+ | IP address | | identifier | +------------+ +------------+ | v +------------+ | IP address | +------------+ What form should the ‘identifier’ take? If it conforms to the domain.TLD layout, then the DNS, can map this identifier to an IP address, without any change. In fact, established meaningful domains and the new identifiers can co-exist. This backward compatibility will ease acceptance. The identifier, rather than the real-world name is to be used in hyperlinks, bookmarks, email addresses and other references. This identifier is guaranteed not to change. (Traditional domain names should also be constant, but aren’t. [ http://news.bbc.co.uk/1/hi/business/7986483.stm]) We can solve (ie delegate to the real name layer) the challenge posed by Internationalized Domain Names, by making identifiers purely numeric. Every computer literate culture has ten digits and differences in representation can be trivially and unambiguously translated. This suggests identifiers of the form dddc-dddc-dddc-dddc.dd-dd, where d is an arbitrary digit 0-9 and c is a check digit. The rightmost group is effectively a numeric TLD, the 16 digit group is the equivalent of the domain name. The check digits in the domain are calculated over each preceding group plus a digit of the TLD with the aim to catch the most common typing errors. The hyphens serve to ensure the string is not interpreted as an IP4 or IP6 number. This form gives us potentially one trillion unique identifiers in each of 10,000 TLDs, with the option to extend the length of either the TLD or the domain. Not all 10,000 TLDs need to be introduced at once. Of course sub-domains can be prefixed to this scheme, within the total length constraints of the DNS. The recommendation is for sub-domains to follow the same dddc convention, implying that the check digit algorithm must be public. There are various refinements to this scheme, such as allowing shorter domains and actioning them, or reserving particularly catchy TLDs for special services, for example tinyurl redirection services. There is ample scope here for innovation and competition. But, in general, neither the TLDs nor the domains should carry any meaning. None is more desirable then the next: no sunrise clause; no string contention; no dispute resolution. No registry is to ‘own’ a particular TLD. This is a strengthening of the earlier observation that registries should compete within one TLD. In this proposal, registries and TLDs are completely divorced. Instead registries will, ideally with competitors, administer the processing of the following classes of real-world names: 1. supra-national institution names 2. globally trademarked names 3. country trademarked names 4. country registered company/charity names 5. government sponsored names 6. names sponsored by other accredited country body 7. names of families or individuals 8. arbitrary, non-registered names 9. … The class of the real-world name and any distinguishing information are to be incorporated in a new DNS record type, to facilitate reverse lookup (from identifier to name, class and information). It is probably not appropriate to burden the DNS with becoming a directory service. The DNS typically gives a single answer to a query, rather than a long list of possible matches. Innovation in directory software and user interface is a corollary of this proposal. Or we can just give it all to Google. In each of the first seven processes, the registry will require adequate proof that the applicant has a legitimate claim to the name. Working through registrars, this means that the registry specifies the nature of the proof, as well as any additional information required to assist the end-user (or software agent) in selecting a name from among a number of potential duplicates. The registry also quality assures the work of registrars. This involves more work and more responsibility than either registries or registrars are currently used to. Instead of merely extracting rent, they actually have to add some value. And the added value is this: Through due diligence certify that the registrant really owns the name. Signal this to the end-user through a community-agreed marker. Value to the end user: easily distinguish global brands, from local companies, from hobby sites. Value to the IP community: despite potential for duplicate names, ownership of brand is clearly flagged. No need for defensive registrations. Existing (defensive) domain name registrations can lapse without risk if the proposed scheme comes into effect and new registrations of traditional domain names cease. Cyber squatting is likely to be much reduced from current levels (It would have to be trademark squatting – an expensive and risky business.) Existing squats will prove valueless over time. Because of this added value to registrants, higher fees are legitimate. However, competition should result in innovation and keep costs down. Innovation could come, for example, in the way registrars obtain with trademark and company information necessary to authenticate a request for a name. Because registries are divorced from TLDs, under this proposal, there is no lock-in and no cost of change – concerns Professor Carlton rightly identified, and lightly dismissed, as problems with the current system. The eighth process allows anyone to register any name with no proof required. Because this involves very little work on part of the registry or registrar, this should be very cheap. These names are clearly flagged as having no kind of certification: it is clear that they are not trademarks or registered names of any kind. So although anyone could register someone else’s trademarked name in this uncertified category – any number of times, in fact – it is clear that these are not trademarks or registered companies or charities or any other accredited entity. That will not stop registrants receiving cease-and-desist notices, or risk being the defendant in an infringement suit. But such a suit should not succeed unless ‘passing off’ was the clear intent. This proposal, then, circumvents the artificial scarcity of domain names, protects the legitimate interests of trademark holders and other certified name owners, yet allows full freedom of expression to the community at large, in any script of their choosing. It meets ICANN’s mission of opening up competition and encouraging innovation, while assuring the stability of the DNS and the internet overall. And I commend it to the community. Henk Bakker Romford UK 17 April 2009