ietf
[Top] [All Lists]

Re: get technical, please? , Re: Trees have one root

2002-07-30 18:25:51
g'day Vint,

"vinton g. cerf" wrote:

ed, how would you suggest to resolve an email address if it returns ambiguous 
results?

Maybe it's because I've spent part of the last week poking
around with synchronous serial protocols (I2C, anyone?) but
I think I may have a useful analogy here.

You can divide serial communication protocols into two
classes, those which included embedded clocking and those
which rely on external clocking. One technique isn't
naturally superior to the other, and in fact there are times
you want to use one or the other, depending upon design
goals, circumstances, etc.

We can think of the DNS service as commonly used today as
having "embedded clocking", in that the knowledge of where
to go to resolve a query is assumed as part of the query.
This is less flexible than having this knowledge outside the
query but it's simpler and does lead to a simpler system. It
also ties everyone together in ways that, because of
admittedly non-technical decisions made outside the scope of
this list, the resulting service is running up against
constraints a lot earlier than might be expected absent such
decisions. Sadly, now that "steampowered.com" is gone I can
wail all I want, but I'm not getting it as long as we all
use the same resolvers (ie. the same "embedded clock"). This
is the only way to guarantee universality of response.

The DNS doesn't *have* to be operated this way. The system
could be implemented to provide different responses to
queries depending upon external factors, such as your
location, load on servers, what you're trying to do, etc. In
such a model queries are still resolved, and with greater
flexibility, for those who can sacrifice universality.

Is this an outrageous idea? Maybe, but this is a technique
that's being used today, for example  in CDN systems to
provide DNS-based query resolution. Users don't even know
it's going on and the world didn't grind to a halt once the
universality of name to address bindings was destroyed. You
ask for a URL, and the server you get to is determined for
you, and yes it may well differ from the one the guy in
Australia gets for the same URL (Personally, I suspect his
includes details of the secret Aussie scramjet technology
which we outside the motherland aren't privy to, but I
digress...  ;-)

This is really a form of "external clocking" of the query
resolution service, and it's done today more than folks seem
to want to admit, both at the level of DNS and at the level
of content filtering (I know a number of folks who have told
me "I couldn't forward that post from work because the
company's mail filter thinks 'shitake mushrooms' are a
scatalogical reference on the disallowed list so wait until
I get home tonight and I'll send it on from the other
machine").

Sure, if you need universality of reference you need a
universal namespace, but I guess I'm starting to believe
that the number of times I could do without, or actively
don't want, such universality is growing enough that I might
want my *next* name resolution service to give me a bit more
flexibility and choice. Hide the nerd knobs behind flipup
panels, but the lack of flexibility in the current system is
getting in the way and I suspect will need to be revamped to
get it ready for the next 100 million users...

                                - peterd


---------------------------------------------------------------------
    Peter Deutsch                       pdeutsch(_at_)gydig(_dot_)com
    Gydig Software


   That's it for now. Remember to read chapter 11 on the
   implications of quantum mechanic theory for time travel
   and be prepared to have been here last week to discuss.

---------------------------------------------------------------------



<Prev in Thread] Current Thread [Next in Thread>