[NANOG] Re: new.net: yet another dns namespace overlay play

Thu Mar 8 10:05:41 UTC 2001

WARNING: This post contains heavy snippage and tongue-in-cheek footnotes.

At 07:38 PM 3/7/2001, you wrote:
>Build a search engine which takes "old" domain name "WWW.CNN.COM" and
>produces URL with 207.25.71.27 in it :)

Great. Now I just have to remember the IP address of my favorite search 
engine. Why don't we simplify it and just remember

>Even better, go to a real search engine and look for "CNN news US
>edition".

I spent a year in the bowels of search engine placement. I can tell you 
that if there's a resource I am going to use repeatedly, on multiple OSes, 
multiple machines, and across any length of time, I don't want to find out 
my favorite search engine purged it's record. Nor do I want to find out it 
has renumbered, and the search engine has the bad data cached. At least DNS 
zones have TTL's. No, that won't happen to the big traffic sites, but it 
sure will happen to the smaller ones. Not only that, but you are swapping 
apples for oranges here; I don't have to remember "cnn.com",  but I have to 
remember "cnn news us edition"?

Not only does this break economical and easy redundancy, it breaks virtual 
webhosting as well. *That* will certainly be a nice kidney punch to the 
ailing IPv4 address space. "Excuse me, Arin? Hi. I need a /20."[1]

>   I'm wondering how people managed to find CNN on TV -- after
>all, CNN ads didn't feature local channel numbers :)

On cable TV, you have ~200 channels. On the internet, you have 4 billion IP 
addresses. The analogy doesn't scale.

Not to mention the fact that you have a channel guide that serves the same 
function as DNS.

>As for removing environment variables and symlinks... hmmm... people who
>built Unix in the first place certainly didn't like these features, and
>replaced them with much more generic concepts in Plan 9 and Inferno.

I see those OSes *all* the time.

Hell, I've seen more copies of Ed Woods "Plan 9 From Outer Space" than I 
have Plan 9.

What you are suggesting is that we remove a universally implemented 
long-held industry standard and replace it with the equivalent of a lot of 
kludgy proprietary hacks- which is what DNS was invented to escape from. 
IMO, this is an astoundingly bad idea. The system as it stands is fault 
tolerant, distributed, universal[2], and easy. You want to replace it with 
something that is more susceptible to human error, more likely to break 
during network renumbering[4], more expensive in both time and money, more 
wasteful of dwindling IP space, and less functional overall?

When your car gets a few scratches on the hood, do you junk it and walk 
everywhere?

At 07:43 PM 3/7/2001, Edward S. Marshall wrote:
>On Wed, Mar 07, 2001 at 07:04:05PM -0800, Ben Browning wrote:
> > For some reason, I can't see CNN broadcasting "Come visit our website, at
> > 207.25.71.27 or 207.25.71.28 or 207.25.71.29 or 207.25.71.30 or 
> 207.25.71.5
>
>"Find us at AOL keyword 'CNN'".

...at which point AOL Keywords will become even more spammed than domain 
names, and more broken than DNS.

>Please, explain to me how DNS differs from any other resource location
>(or perhaps "association") system? Examples that immediately spring to mind:
>
>- Address books
>- URL Bookmarks
>- Altavista
>- Google "ad words"
>- dmoz.org
>- doubleclick.net banners

All of these currently point to a distributed, authoritative resource 
system. They cannot be relied upon to be authoritative in and of 
themselves. I have seen pages on Altavista, for example, that have been 
defunct for *over a year* and they are stiull listed. Despite numerous 
attempts to get them unlisted.

>- LDAP

"Unlike existing database systems, LDAP is not designed to hold many 
hundreds of thousands of entries. It might be best to think of LDAP as a 
hierarchically organized lightweight database. An LDAP server may use a 
small embedded database to contain its information for faster access, but 
it's nothing like the large commercial databases such as Oracle, Sybase, 
DB/2 or SQL Server. "
       http://linuxworld.com/linuxworld/lw-1999-07/lw-07-ldap_1.html

>- AOL keywords

No real difference, aside from having 1/3 the amount of address space that 
the Big Three TLD's have. Then again, do we really want to standardize on 
an AOL product that remains wholely under their thumb?

>- ARIN allocations

...and apnic, and ripe. Aside from a layer or two on the ol' OSI model?

>- akamai

Difference? Explain the similarity?

>- BGP updates

My router automagically remembers ASN's it talks to. I have a phone book 
for a reason.

>Some, you can register with for free. Some, you have to pay for. All act as
>means to locate resources (URLs, email addresses, etc). The proliferation of
>these is, IMHO, an indictment of DNS as a resource locator; obviously, it
>isn't usable or general enough to serve the needs of today's Internet, or it
>(and its implementors) would have kept up.

A Swiss army knife with a hammer attached to it will never equal a true 
hammer. The only thing that even comes close to DNS in terms of the niche 
it fills is the AOL keywords thing, and even that already uses DNS as a 
backend.

Bottom line:

Just because it ain't perfect don't mean it ain't the best solution.

~Ben, as always, speaking for himself

[1] Which, I believe, is a very special level of Hell reserved for people 
like Hitler, Stalin, and Barney the Dinosaur
[2] Although I do agree that unicode support would be nice.[3]
[3] Man, the skr1pt k1dd13z domains would go nutso with that one
[4] This is the floor right above the one referenced in [1]

#
# Comment: You may have to Ctrl-C out of the footnote loop above
#
---
    Ben Browning <benb at theriver.com>
       The River Internet Access Co.
              Network Operations
1-877-88-RIVER  http://www.theriver.com