The global domain name service is obsolete.

Posted 15 Jun 2000 at 21:08 UTC by logic Share This

ICANN, the Internet Corporation for Assigned Names and Numbers, has opened up a public comment period for their Domain Name Supporting Organization's proposal on the introduction of new top- level domains. In this essay, which I've submitted to their comment system, I try to outline my assertion that DNS, on a global scale, has

outlived it's usefulness in lieu of technologies more appropriate for the massive level of growth that the Internet has seen since DNS was first put forward as a means of hierarchal addressing and location.

"The global domain name system is obsolete."

That's a pretty bold statement to make. Think about it a moment, mull it over: "DNS is a relic of a set of problems which have been solved in more appropriate ways".

I don't think that assessment is far off the mark. Right now, you have an infinite number of organizations and individuals fighting over a finite namespace. As we're beginning to see, that approach is failing miserably. And yet, for some reason, we're clinging to this system which in the Internet of today is no longer needed.

There have been a number of recommendations to address this, from ICANN suggesting that additional top-level domains be added, to some which have suggested an overhaul of how naming is handled, and brush up against the issue. But these all miss the fact that, globally, we already have an addressing system; it's at the heart of everything, has no political or legal baggage, and is something that many people these days take for granted: IP addresses.

"But wait," you might say, "how do we find a particular organization or individual website?" The days of when you could simply type in a company name and place a ".com" on the end of it are quickly departing, which means you need to bookmark websites that you've previously found interesting, and that you need search engines to locate websites that you haven't found yet. A mnemonic name is really only a convenience for marketing materials at this point, and with the shortage of names, the mnemonic value is quickly lost too as more and more "names" begin looking and sounding very similar (and this doesn't even address the problem of DNS names not handling international character sets, meaning that names are essentially locked into the letters A through Z, and the numbers 0 through 9).

"But wait," you might say, "how do we handle email in a DNS-less world?" The same way; LDAP-based people registries are a far more efficient way of locating people you're trying to find (rather than blinding trying addresses), and addressbooks have long been the only way for most people to really manage the addresses of people they know. When you only need to type in "emarshal@[]" once, and can reference it from then on as "Edward S. Marshall", "Ed", or "Sysadmin", the original address quickly becomes irrelevant as the mnemonics have once again been lost, replaced by a mnemonic much more suited to the individual user's needs.

"But wait," you might say, "how do we conveniently reference other systems?" Easily: enterprise-level DNS. Managing namespaces internally makes even more sense with the trends leading toward non-routable IP addresses for numbering an internal network. Just like each person references other people by privately-managed addressbooks and bookmarks, locally-managed DNS is a perfect means to handle intra-organizational addressing. Most organizations already do this, using DNS or WINS to manage the naming of their systems independant of the rest of the world. These names become irrelevant when the systems they refer to are private or unreachable.

"But wait," you might say, "how do we perform geographically-diverse load balancing?" Load-balancing technology has already reached the maturity level necessary for this, and is in heavy use today. DNS is a poor way to perform robust load balancing; it cannot handle anything other than evenly distributed balancing, and cannot even guarantee that. Use of an IP-level load-balancing product is a far more appropriate approach to the problem given what is available today. In a web-based application, the problem is solved in an even easier manner through the use of redirection. DNS, while a good "first try" at solving the load balancing problem, really fails to solve it as well as current technologies.

"But wait," you might say, "how do we renumber without an impact on our availability?" A similar question, with equally damaging ramifications, is "how do we rename after a trademark dispute takes our domain name from us?" The same problem looms here, and unless there is an exact mapping of trademark law to domain management, there will always be a risk of having your name taken. Just like the post office assigns an address to a business, ARIN and similar organizations (by proxy through your ISP/NSPs) assign your addresses to you. These addresses are determined by your location; for the post office it's a geographical assignment, and for ARIN/RIPE/etc, it's a topological assignment. Either way, changing addresses should always be well-considered before doing it, and updating registries (business listings, phone books, search engines, people directories, etc) should be an important part of that consideration.

In the end, the need for DNS to be our global means of indexing content has outlived its usefulness. Local DNS servers can be used for the need of local addressing, and IP addresses serve the need for global location. Bookmarking and address registries have replaced the need for mnemonic names to refer to Internet systems.

The global domain name system is obsolete. Please don't artificially prolong the life of a system that is already dead.

That's the full text as posted. I'd expect people to disagree with me on this one, and I hope Advogato will provide a good forum for this discussion, given the nature of its participants.

Few random thoughts in a random order., posted 15 Jun 2000 at 23:03 UTC by Iain » (Master)

A few comments that I can think of - in no order other than the one I write them in.

  1. Changing IP address happens more often than trademark dispute.
  2. Not everyone uses bookmarks, and the only time I use bookmarks is when I'm in a deep directory that I'd never remember. Not when I'm at's front page.
  3. People cannot remember more than 8 numbers or something like that. IPV4 is 12, IPV6 is what?
  4. People remember names more than numbers.
  5. Even if now you can't just stick .com on a new company's name, you can do it to most established companies. You would want to scrap that because newer companies have to have more exotic names?

Just some thoughts, but I'm probably way off, cos I can't read things on the screen very well...yadda yadda excuses....

(I'd still like my idea of a truely commerical free tld, no companies, no adverts. What .org was probably like in the old days :))

IP Level Load-Balancing, posted 15 Jun 2000 at 23:27 UTC by highgeek » (Master)

I must say that sometimes you do need load-balancing that the IP Level routing can not handle very well. You could get some Gigabit switch to potentially handle this, but it would have to be some huge iron horse. You would need at least two of them to make sure the first one stays alive and that could quite expensive quickly.

For HTML pages and such I would totally agree with you, but with Software and Multimedia, I must say that it would be a hard feat to create a load balancing device (without RR DNS) to handle more then 200 MBits of traffic to a cluster of 20 or so machines.

If anyone has any ideas on how to do this, I would like to hear them. It is definately a problem that strikes my interest. :-)

uhh, no, posted 16 Jun 2000 at 00:45 UTC by graydon » (Master)

the assumption that [# of DNS trademark disputes] >= [# of network relocations] x [# of referenced hosts on the relocated network] is absurd.

the cultural effect of having a standard, kind of kooky but nonetheless useable global name->number map is significant. you need to provide years of software modifications, bridge mechanisms, education and arguments in order to facilitate a transition to LDAP. maybe some day we can declare DNS "dead", but not today.

I have a dream .. , posted 16 Jun 2000 at 01:02 UTC by joey » (Master)

This is the main failing you identify with the current DNS system:

I don't think that assessment is far off the mark. Right now, you have an infinite number of organizations and individuals fighting over a finite namespace. As we're beginning to see, that approach is failing miserably. And yet, for some reason, we're clinging to this system which in the Internet of today is no longer needed.

It seems to me that the answer is to use the DNS system as it is really meant to be used, not the form it has been perverted into in the past 5 years. Namely, use deeper hierarchies.

Remember, DNS is a hierarchical system. In an ideal world, advogato would be, slashdot would be Microsoft would be; apple would be apple.{software,hardware} My homepage would be under * And of course * would be chock-full of every porn site on the net today. If all you're interested in is porn, or the free software community, you then have easily delinitated places to look.

This is not a limited namespace, it is near-infinite if it's made deep enough.

I think what we need to do is make one new TLD, and no more. Take that TLD, and load it down with very explicit rules. No top-level domains may exist. People/organizations may register to be responsible for administering a hierarchy under a top-level domain, but make very clear and explicit rules about what they must do to prevent abuse by them locking out people who deserve to get a place in the hierarchy. (Or perhaps allow multiple organizations to administer the same hierachy, like top level dns is handled now). Allow those who are registrars for a part of the hierarchy to set up web sites indexing the part of the hierarchy they control, so there is some incentive for them.

Why have we so flattened this hierarchy to a system where you take your company's name and slap a .com on the end? Seems to me it's because companies came to the net and realized that for marketing reasons, they wanted a simple keyword to be their address. And as things were set up, they had no difficulty getting what they wanted.

For this new TLD to work, we would somehow have to sell it to the companies, make them see there is benefit in organization.

I don't think this utopia will ever happen, but I can always dream..

problems with tree heirarchies, dislike of IPs, posted 16 Jun 2000 at 03:35 UTC by splork » (Master)

Tree heirarchies are useful, just not to humans. People do not remember trees very well (especially when they start containing many many branches). Our top level domains today are effectively "aol keywords" to the masses because they're easy to remember and communicate on non-digital medium.

Trees also bring the problem of recategorization and management. Who decides what goes where in a tree? When should something be moved or allowed to exist in multiple places within the tree?

The problem with the suggestion of IP addresses being our names is that our IP addresses can't be carried with us when we switch ISPs, move, etc. Cell phones numbers have a similar problem (but that's an off topic rant...).

Is anyone familiar with IPv6 developments? I thought at one point there were going to be "entity" addresses (fixed, permanently assigned; 48-64 bits or so) that could always be addressed in the way the author was hoping current IP addresses could be used. Is this true? Doesn't this make for routing nightmares? Ah well, it'll be a -long- time before or if IPv6 is ever widely adopted.

Eddie!!!, posted 16 Jun 2000 at 17:14 UTC by Toby » (Master)

Ok, maybe not an appropriate "response", but I used to mark this guys assignments. I could tell you guys stories... but I won't. It's nice to see one of my old buddies on here. Sorta makes me feel all old...

Anyways, nice essay, interesting viewpoint. Allow me to address or reply to some of the points that you made. First and foremost, the act of finding a website and/or email address could be solved in a number of manners. LDAP could certainly be among them, along with search engines, and other databases to keep track of things. Administration of these things can also be done any number of ways. Of course, for any of this to start taking off, there will have to be software out there that is both easy to understand and use, and that will scale for the next 10 years (at current exponential rates).

About your enterprise-level DNS (or whatever naming system), there are certain "problems" with this. The worst one I can think of, is that this will only really be a solution to use for things internal and within an organization. Once you get to things like freshmeat, /. and other things which wish to have a global impact and presentation, it makes sense to have these be refered to by the same name/acronym/etc by everybody. Otherwise, if you switch to a new job (congrats on yours), you may have to relearn how to connect to your favourite sites, as the local administration may have renamed things for some particular reason.

This is not an unsolvable problem, but enterprise-DNS (or other naming systems) do tend to require a certain amount of maintenance, maintenance that is sometimes hard to justify to management. (Yes, I still work in an education environment...)

Yikes, there is a subject that should have a whole new thread. In today's world, IP numbers are transient at best. I've always wondered why there was not a 3 level architecture to translating DNS names to IP numbers. The scheme I have in mind would work something like: everybody can request a unique identifier, make it 128bits long. This is the number that identifies you. If somebody wishes to get a hold of you, and they happen to know this number, they can. So, telephones, IP, whatever way. Take the IP version of this for now. You would have something along the lines of DNS (Yes, but for this DNS s/would work ok), to translate this 128bit number into your current "IP" (IPv4/6, etc) address. At this point, depending on how fast this translation happens, mobile IP becomes a reality (actually location independant IP), as well as sort of solving the renumbering problem.

Also, since there is a "fixed" level (level 2), we can build on the other side of this, and make a DNS, LDAP or whatever nameservice/index you wish on the other side. I'm not sure if I'm making sense here, but it would become a 2 layered translation. The first from something possibly changing (lookup in more than one possible place, such as local DB, DNS-type, LDAP, etc) and enterprise specific, to the second which is a "static", globally unique identifier, to something semi-dynamic, which is your final destination.

Note: would you not have liked to give somebody (maybe your bank, or whoever) a simple 128bit number, so that they can always reach you, no matter where you are? Yes, you would still need to "update" your information to make the final routing of whatever (telephone, mail, email, etc) complete, but this would hopefully be done in an easy and painless matter.

Note 2: Yes, there are privacy issues here, and I'm not even gonna try to address them.

everything is deeply intertwingled, posted 20 Jun 2000 at 05:07 UTC by jwz » (Master)

Joey wrote:

In an ideal world, advogato would be, slashdot would be Microsoft would be; apple would be apple.{software,hardware}
I don't think this utopia will ever happen, but I can always dream..

Wow, your utopia sounds like hell to me! DNS may be hierarchical, but the real world isn't, and I for one am very glad of that. You think slashdot is "news", but measured by volume, it's "community". (And now they're a commercial enterprise that makes money selling ads, doesn't that make them a .com?) Some people think my home page is entertainment, others think it's software. New Glimmer is a floor wax and a dessert topping.

``Intertwingularity is not generally acknowledged -- people keep pretending they can make things deeply hierarchical, categorizable and sequential when they can't. Everything is deeply intertwingled.''

-- Ted Nelson

The problem with DNS may be that it is a hierarchy but the real world is not. Or it may be that DNS is a hierarchy but everyone treats it as a flat space. But regardless of how poorly it fits the world, I think it's here to stay. It will always be subsidiary to trademark law.

Perhaps DNS should follow trademark registrations more closely: perhaps the domain-registration process should involve some kind of automated trademark search.

Germs of interesting ideas, posted 25 Jun 2000 at 06:11 UTC by lilo » (Master)

I'm seeing several good points. IPv4 addresses are too transient. IPv6 might be used for a fixed 'entity' address system. DNS is hierarchical but people aren't; people don't remember the names well when the trees are too deep, and more importantly, who decides the tree and branch in which to place an entity?

That last question seems really important. Who assigns IPv4 addresses? The ISP, and those IP addresses are relatively transient in that you can change providers. But who assigns fixed IPv6 'entity' addresses? Some single organization. Who assigns you into your tree position? Within that tree, same answer.

The real problem is this: When you insist on a single namespace, some single somebody generally has to control each section of it. In the case of the root server, that can be a government or quasi-governmental entity; in the case of the 3 main TLD's that can be a cartel such as was recently created.

When a single entity maintains a resource, that resource is potentially a bottleneck and you can accrue power and money by controlling it. So as long as we have a single namespace or a single hierarchy, we have problems.

It's worth considering that in many many cases the domain names and the email usernames we use aren't especially memorable. If my acquaintance foo has, and he's not a particularly cheerful person, his domain name doesn't jog my memory. And if my friend Fred has, ditto. It would help if there were a larger namespace available for people to use, but how much of the .com namespace is taken up by squatters? It's difficult to define what a squatter is, but easy to see the results. Get a dictionary, a browser, and a copy of your favorite 'whois' binary and see how many English words are taken, and how many of those domains actually have some meaningful content on a web page.

Where does this leave us? I'd love to see a completely arbitrary-looking but relatively permanent naming convention based on long public-keys, with peered domain name servers operated on a purely informal basis. Let the public use the peer services they select and decide which ones work. One might be surprised how quickly that would shake out into a reliable system, and no one could possibly have a lock on such a system. Everybody picks their own key pair and the signed, dated domain name records tell you what their preferred nickname is. Let companies take people to court if they're abridging trademarks in commerce. Use something like LDAP to maintain the records.

Technical types, the sort of people who like to innovate, could start using something like this any day. You could start using it for email, and get the added advantage of being somewhat invisible to most junkmail. 8)

Just a few ideas off the top of my head. Interesting subject matter.

DNS, ala Usenet, posted 30 Jun 2000 at 14:35 UTC by logic » (Journeyer)

That's actually an interesting idea, one I hadn't considered. Approaching resource location services from a peering point of view, rather than centrallizing it, could be interesting.

The problems I can see are similar to the problems that Usenet or the global routing tables have; at it's current level of volume, only those with relatively deep pockets can run a complete news server (or core, well-peered router). As well, the possibility of abuse is extremely high; how do you prevent the injection of bogus names and namespace collisions (for example, as in IRC), without centrallizing some kind of authentication and monitoring?

It would be interesting to see the idea fleshed out, though. :-)

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!

Share this page