Current Meeting Report

2.1.15 Internet Resource Name Search Service (irnss) Bof

Current Meeting Report


Raw notes, thanks to RL 'Bob' Morgan

Klensin (JK) introduction:

One of the general purposes of the BOF and any future effort is to offload non-DNS, non-IDN stuff from IDN WG since that group is a magnet for such stuff. But, more generally, it is an effort to figure out ways to do naming-related things for which there is a clear demand, including language-based internationalization, for which the DNS is not suitable.

"classification systems" in general are faceted; hence uniqueness, reserved words, hierarchy
can uniqueness and registries (arbiters) be avoided?
can registered servers be avoided?
complex client-side solutions (some proposed in IDN) are suspect

comment (from whom?)
distinguish operations that require human user interaction from those that don't
those that do are "ratbait", ie attractive nuisances


Leslie Daigle (LD), SLS document
problem statement
"provide Internet service naming that":
all languages, scripts
non-uniqueness, global
independent of network element names
can be used as basis for naming in other protocols and software
note layer 3 as distinct layer

how granular can location be and be interoperable?
is exact match only OK?
is a small list of taxonomies good? is it do-able?
is Unicode sufficient?

JK: success leads to uniqueness collisions
so desire for uniqueness leads to regulation and/or monopolies

JK: books-on-shelves analogy
life is fuzzy
DNS is "downward-looking" since can computers can be non-fuzzy
need pointer to shelf, not pointer to library
we like generality and presentation of complexity to users and they freak out, want simple well-defined answers

Micheal Mealling (MM): this is part of question of "layers" and "on top of"

Harald Alvestrand (HA): meaning of "global"?
like DNS, same answer no matter who you ask?
or, can get to it from anywhere, but there can be many?

MM: thinking of latter, can have many speakers
but it's OK to ask many at once

are objects of interest "data objects" or "names"?

MM: main deal is services that user use

JK: would prefer to preserve traditional service/port model
are returned thingies DNS names or URIs?
URIs are scary due to potential for recursion

is this another naming system on top of DNS just for apps?
these will leak into end-user experience
LD: this is layer 2, which is defined for apps
layer 3 is designed for end-users

PH: not just languages and scripts but cultures
where eg "industrial classification" is typically first-world-specific
esp "what kind of person are you"

JK: reacting to prevalence of trademark lawyers and such
hoping to move questions to localities rather than global trademark lawyer: it's too late to hope that TM law can be made local
US cybersquatting law is international disaster

LD: but point is to do whatever TM law does, instead of circumventing it

TML: ICANN UDRP is a start on this but industry codes are a dead-end

MM: but maybe better than nothing at all?

librarian: classification systems viewed as non-robust

MM: client-side

Ted Hardie (TH): nice to hear that layer 2 is spose to be for apps only but almost everything in SLS doc is user-oriented

LD: revised text appreciated

JK: this is broadening of options available to app
where apps often just die in the face of unusual behavior
gotta get the caching right too of course

TH: SLS stuff isn't cacheable at the moment, a problem
is a fuzzy-search service infra or application itself?
where distinction is based on determinism of answers?
(false fire alarm happens ...)

bearded fellow in purple shirt:
adding integrity and protection as afterthoughts is really bad

MM: hoping to use signed XML, only this isn't XML
not as bad as DNS in terms of signed delegation

BF: that's because you haven't thought about delegation yet

PH: granularity of location issues very different between companies and people
eg people probably want to be postal addrs ...
meta-issue is what's in which doc

MM: SLS is just about layer 2

JK: possibility of supporting multiple locations

Dave Crocker (DC): this is nexus of high-need, high-benefit, high-difficulty and low-understanding

MM said: this is one of many layers, and all must be present
yes, all must be present, this guarantees failure
so, have to have many smaller steps

LD: OK, SLS is a concrete proposal ...

MM: well, companies are deploying stuff now, and are interested and are making money, so we do have some experience

JK: is layering just an excuse to punt to undefined layer?

MM: well, it's important to talk about related layers
so well-defined layer can be described

JK: but if so much that people want to do is punted to layer 3 ...

Nico Popp (NP): facets are cool because they allow phased granularity
deployed systems use country-based geog now, but can get finer

KM: uniqueness enforced by service provider is foolish


Nico Popp and Yves Arrouye (YA)
advocating keyword systems
many in use making value today
eg AOL, Realnames, many in Korea/China
meetings have been had with operators of these, about requirements
#1: uniqueness of service descriptions?
#2: support all scripts
#3: multiple data repositories
#4: URIs as returned items
all systems have chicken-and-egg
nobody registers or uses until there's lots of data in it
so signing up a lot of early registrants is pragmatic
may not reflect preferred final architecture

CH: how do you do search across providers?

NP: CNRP has referrals ...

CH: input strings as search expressions? this explodes

YA: well, current systems just take strings

PH: keyword systems got bad rep by scaring people into registering names-as-concepts (eg automobile) before someone else did

NP: well, it's pragmatic, now we have rules and elaborate review

JS: why does China have multiple-database requirement?

NP: per-country level of databases, at least

JS: have to distinguish user requirements and business requirements

MM: avoided "keyword" term in SLS doc, since it means different things, eg with CH above

JK: kw system has lots of words, user chooses some for query systems such as these are different

Karen Liu, from Chinese keyword provider:
users should choose the provider they think is best, not govt

Lib: works if vocabulary is very controlled, otherwise not

JK: these systems have name strings and other facets
different providers have same facets, different name strings
slide is displayed showing keywords in multiple scripts/languages
are these "trademarks"? sure, among others

DC: talk about independence and choice
WWW works with URLs, which unify services that stay different
and domain names, which are centralized
having name depend on provider is a lose

PH: user discovery of directories

users ask each other "how did you find that?"

Google is popular because it's easy to choose one thing

users also see advertising, also stuff is built in to software they get ...

so, how about a directory description format so directories could describe themselves and users could describe ones they've used

should describe use of facets

user should be able to name/transmit dir bundle including ratings/annotations about relevance/areas

is this IETF work?
* could do format, bundling, no protocol
* no need to start until other sublayers are defined
* propose not thinking about this yet

NP: CNRP already does this ...
PH: sure, could choose that
NP: meta-directory of these services could exist too
PH: university librarian could put this together, with recs
YA: would also be useful to capture search as useful search to be handed to others


final questions:

is IETF defining level-2 service good-to-do? 20 or so
is IETF defining level-3 service good-to-do? almost none

ietf-irnss, signup at


User discovery of directories
IRNSS - General
Service Lookup System