[ih] DNS History

[John Day]

X.500 in essence killed itself.  (Hoyt Kesterson will kill me for 
saying so . . . )

As usual a number of things came together to kill X.500.  For the 
most part, it was outside the rest of the OSI train wreck.  Like most 
OSI application protocols, it tried to do too much with no real 
approach to reasonable subsets.  (We tried to impress this idea on 
the Europeans from the early 80s on but they would have none of it.) 
The primary problem was that the directory wasn't an application 
protocol in the usual sense in the first place.

First of all, a "directory" was suppose to be something that *only* 
did application name to address mapping.  This is how it is defined 
in the Naming and Addressing Part of the OSI Reference Model 
(7498-3).  (DNS does something entirely different.  Initially, DNS 
translate a string representation of an IP address to a bit string 
representation.  Today, it has morphed into something in between that 
isn't really a clean separation of application name and network 
address and hence not as rich as is needed for a complete 
architecture.   See Shoch and Saltzer.  Grapevine was the first 
attempt to get it right.)

However, X.500 couldn't just do *that* one thing.  They had to make 
it directory for everything.  In essence, X.500 tried to be an 80s 
concept of Google and a directory all rolled into one, when they 
should really be two different things.  In fact, the early drafts had 
something called a descriptive name that was indistinguishable from a 
query.

X.500 was done at the height of the RPC fad.  *Everything could be 
done with RPC!*  Request/Response is everything.  One of the more 
foolish ideas to sweep through computer science, even then.  (I made 
more than a few of them unhappy when I pointed out their wonderful 
new idea was nothing more than COBOL coroutines.) X.500 was done by 
the same crew that did X.400 who were madly in love with 
client/server and RPC and none too swift.  They would have pages of 
syntax definitions (in ASN.1) labeled "Formal Description of the 
Protocol."

When you tried to explain to them that there was more to specifying a 
protocol than just the syntax, you would get blank stares.  When you 
pointed out that you needed to specify the *procedures* what to do 
when a PDU arrived, the still looked at you blankly.  I remember a 
big meeting that Jack Veenstra and I had with John White, PARC and 
the rest of the X.500 crew. They thought the names of the attributes 
in X.500 *were* the definition.  That was when I pointed out that I 
could use the letter "Z" as a value in every field in their protocol 
and it would be conformant.  "But that is not what we meant!"  But it 
was what you specified.  ;-)

Their RPC is everything model sort of broke down as well, when they 
realized querying the directory wasn't the only thing that had to be 
done.  There had to be directory updates as well and they would be 
really bad if they had to request changes rather than be notified of 
them.

It was a classic case of generalizing off a model that was in fact a 
special case.

I was always surprised that it lasted as long as it did.

Take care,
John

At 12:58 -0800 2010/03/08, Richard Bennett wrote:
>X.500 was a much broader-reaching directory service, whereas DNS was 
>a simple name-to-address mapper. Companies such as Novell did their 
>own directory services, and X.500 never took off because of the 
>skullduggery that killed OSI. John Day's Patterns in Network 
>Architecture covers some of the drama.
>
>On 3/8/2010 12:31 PM, Craig Partridge wrote:
>>>First, in terms of the RFC system, where are the comments themselves?  Were
>>>they hard-copies that no longer exist, or mailing lists that have been
>>>tucked away somewhere?  Is there any correspondence left (for DNS related
>>>RFCs) or has it all been lost?
>>>
>>There was no formal comment system (nor is there now).  But there were lots
>>of comments on drafts on various mailing lists.   For DNS issues the
>>archives of the namedroppers list is probably your best place
>>(http://psg.com/lists/namedroppers and kudos to Randy Bush for bringing it
>>up)
>>
>>
>>>Second, does anyone have or know where to find details about the
>>>debates/conversations that took place leading up to RFC 1591 and what
>>>appears to be a "compromise" between generic and ccTLDs?
>>>
>>RFC 1591 is awfully late -- most key technical issues, as I recall, were
>>determined when RFC973 came out.
>>
>>
>>>Third, it is not entirely clear to me exactly why DNS was engineered in
>>>place of X.500.  It is my understanding at this early point in my research
>>>that OSI standards seemed inevitable at one point, and sources have told me
>>>that DNS was designed to get something out the door quickly (presumably
>>>something that *wasn't* X.500).  Was X.500 simply based on an old paradigm
>>>(white pages / old telecom) and seen as a bulky and slow alternative?  When,
>>>and with whom, was the actual decision made to ditch X.500 altogether?  This
>>>part of the story goes a long way to explaining why everyone in the world
>>>doesn't have a unique identifier.
>>>
>>I have my theory on that subject -- I'll send you the relevant paper I wrote
>>on the history of email, there's a brief discussion.
>>
>>Thanks!
>>
>>Craig
>>
>
>--
>Richard Bennett
>Research Fellow
>Information Technology and Innovation Foundation
>Washington, DC