[Search] [txt|pdfized|bibtex] [Tracker] [Email] [Nits]
Versions: 00 01 02                                                      
Internet Draft                                                M. Duerst
<draft-duerst-dns-i18n-00.txt>                     University of Zurich
Expires 10 June 1996                                   10 December 1996

                  Internationalization of Domain Names

Status of this Memo

   This document is an Internet-Draft.  Internet-Drafts are working doc-
   uments of the Internet Engineering Task Force (IETF), its areas, and
   its working groups. Note that other groups may also distribute work-
   ing documents as Internet-Drafts.

   Internet-Drafts are draft documents valid for a maximum of six
   months. Internet-Drafts may be updated, replaced, or obsoleted by
   other documents at any time.  It is not appropriate to use Internet-
   Drafts as reference material or to cite them other than as a "working
   draft" or "work in progress".

   To learn the current status of any Internet-Draft, please check the
   1id-abstracts.txt listing contained in the Internet-Drafts Shadow
   Directories on ds.internic.net (US East Coast), nic.nordu.net
   (Europe), ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific

   Distribution of this document is unlimited.  Please send comments to
   the author at <mduerst@ifi.unizh.ch>.


   Internet domain names are currently limited to a very restricted
   character set. This document proposes the introduction of a new
   "zero-level" domain (ZLD) to allow the use of arbitrary characters
   from the Universal Character Set (ISO 10646/Unicode) in domain names.
   The proposal is fully backwards compatible and does not need any
   changes to DNS.

Table of contents

   1. Introduction ................................................... 2
     1.1 Motivation ...................................................2
     1.2 Notational Conventions .......................................3
   2. The Hidden Zero Level Domain ................................... 3
   3. Encoding International Characters .............................. 4

                          Expires 10 June 1997          [Page 1]

Internet Draft    Internationalization of Domain Names  10 December 1996

     3.1 Encoding Requirements ........................................4
     3.2 Encoding Definition ..........................................4
     3.3 Encoding Example .............................................6
     3.4 Length Considerations ........................................7
   4. Usage Considerations ........................................... 7
     4.1 General Usage ................................................7
     4.2 Usage Restrictions ...........................................7
     4.3 Domain Name Creation .........................................8
     4.4 Usage in URLs ................................................9
   5. Alternate Proposals ............................................10
     5.1 The Dillon Proposal .........................................10
     5.2 Using a Separate Lookup Service .............................11
   6. Generic Considerations .........................................11
     5.1 Security Considerations .....................................11
     5.2 Internationalization Considerations .........................11
   Acknowledgements ..................................................11
   Bibliography ......................................................12
   Author's Address ..................................................13

1. Introduction

1.1 Motivation

   The lower layers of the Internet do not discriminate any language or
   script. On the application level, however, the historical dominance
   of the US and the ASCII character set [ASCII] as a lowest common
   denominator have led to limitations. The process of removing these
   limitations is called internationalization (abbreviated i18n).  One
   example of the abovementioned limitations are domain names [RFC1034,
   RFC1035], where only the letters of the basic Latin alphabet (case-
   insensitive), the decimal digits, and the hyphen are allowed.

   While such restrictions are convenient if a domain name is intended
   to be used by arbitrary people around the globe, there may be very
   good reasons for using aliases that are more easy to remember or type
   in a local context. This is similar to traditional mail addresses,
   where both local scripts and conventions and the Latin script can be

   There are many good reasons for domain name i18n, and some arguments
   that are brought forward against such an extension. This document,
   however, does not discuss the pros and cons of domain name i18n. It
   proposes and discusses a solution and therefore eliminates one of the

                          Expires 10 June 1997          [Page 2]

Internet Draft    Internationalization of Domain Names  10 December 1996

   most often heard arguments agains, namely "it cannot be done".

   The solution proposed in this document consists of the introduction
   of a new "zero-level" domain building the root of a new domain
   branch, and an encoding of the Universal Character Set (UCS)
   [ISO10646] into the limited character set of domain names.

1.2 Notational Conventions

   In the domain name examples in this document, characters of the basic
   Latin alphabet (expressible in ASCII) are denoted with lower case
   letters. Upper case letters are used to represent characters outside
   ASCII, such as accented characters of the Latin alphabet, characters
   of other alphabets and syllabaries, ideographic characters, and vari-
   ous signs.

2. The Hidden Zero Level Domain

   The domain name system uses the domain "in-addr.arpa" to convert
   internet addresses back to domain names. One way to view this is to
   say that in-addr.arpa forms the root of a separate hierarchy.  This
   hierarchy has been made part of the main domain name hierarchy just
   for implementation convenience. While syntactically, in-addr.arpa is
   a second level domain (SLD), functionally it is a zero level domain
   (ZLD) in the same way as "." is a ZLD.

   For domain name i18n to work inside the tight restrictions of domain
   name syntax, one has to define an encoding that maps strings of UCS
   characters to strings of characters allowable in domain names, and a
   means to distinguish domain names that are the result of such an
   encoding from ordinary domain names.

   This document proposes to create a new ZLD to distinguish encoded
   i18n domain names from traditional domain names.  This domain would
   be hidden from the user in the same way as a user does not see in-
   addr.arpa.  This domain could be called "i18n.arpa" (although the use
   of arpa in this context is definitely not appropriate), simply
   "i18n", or even just "i". Below, we are using "i" for shortness,
   while we leave the decision on the actual name to further discussion.

3. Encoding International Characters

                          Expires 10 June 1997          [Page 3]

Internet Draft    Internationalization of Domain Names  10 December 1996

3.1 Encoding Requirements

   Until quite recently, the thought of going beyond ASCII for something
   such as domain names failed because of the lack of a single encom-
   passing character set for the scripts and languages of the world.
   Tagging techniques such as those used in MIME headers [RFC1522] would
   be much too clumsy for domain names.

   The definition of ISO 10646 [ISO10646], codepoint by codepoint iden-
   tical with Unicode [Unicode], provides a single Universal Character
   Set (UCS).  A recent report [RFCIAB] clearly recommends to base the
   i18n of the Internet on these standards.

   An encoding for i18n domain names therefore has to take the charac-
   ters of ISO 10646/Unicode as a starting point.  The full four-byte
   (31 bit) form of UCS, called UCS4, should be used. A limitation to
   the two-byte form (UCS2), which allows only for the encoding of the
   Base Multilingual Plane, is too restricting.

   For the mapping between UCS4 and the strongly limited character set
   of domain names, the following constraints have to be considered:

   -  The structure of domain names, and therefore the "dot", have to be
      conserved. Encoding is done for individual labels.

   -  Individual labels in domain names allow the basic Latin alphabet
      (monocase, 26 letters), the "-" inside the label, and the ten dec-
      imal digits in all but the initial position. The capacity per
      octet is therefore limited to somewhat above 5 bits.

   -  There is no need nor possibility to preserve any characters.

   -  Frequent characters (i.e. ASCII, alphabetic, UCS2, in that order)
      should be encoded relatively compactly. A variable-length encoding
      (similar to UTF-8) seems desirable.

3.2 Encoding Definition

   Several encodings for UCS, so called UCS Transform Formats, exist
   already, namely UTF-8 [RFC2044], UTF-7 [RFC1642], and UTF-16 [Uni-
   code]. Unfortunately, none of them is suitable for our purposes. We
   therefore use the following encoding:

                          Expires 10 June 1997          [Page 4]

Internet Draft    Internationalization of Domain Names  10 December 1996

   -  To accommodate the slanted probability distribution of characters
      in UCS4, a variable-length encoding is used.

   -  Each target letter encodes 5 bits. Four bits are used as data
      bits, the fifth bit is used to indicate continuation of the vari-
      able-length encoding.

   -  Continuation is indicated by distinguishing the initial letter
      from the subsequent letter [alternative: distinguish leading let-
      ters from final. Pros? Cons?].

   -  Leading four-bit groups of binary value 0000 of UCS4 characters
      are discarded, except for the last TWO groups (i.e. the last
      octet).  This means that ASCII and Latin-1 characters need two
      target letters, the main alphabets up to and including Tibetan
      need three target letters, the rest of the characters in the BMP
      need four target letters, all except the last (private) plane in
      the UTF-16/Surrogates area [Unicode] need five target letters, and
      so on.

   -  The letters representing the various bit groups in the various
      positions are chosen according to the following table:

        Nibble Value   Initial   Subsequent
        Hex  Binary
        0    0000      G         0
        1    0001      H         1
        2    0010      I         2
        3    0011      J         3
        4    0100      K         4
        5    0101      L         5
        6    0110      M         6
        7    0111      N         7
        8    1000      O         8
        9    1001      P         9
        A    1010      Q         A
        B    1011      R         B
        C    1100      S         C
        D    1101      T         D
        E    1110      U         E
        F    1111      V         F

   [Should we try to eliminate "I" and "O" from initial? "I" might be
   eliminated because then an algorithm can more easily detect ".i". "O"
   could lead to some confusion with "0".  What other protocols are
   there that might be able to use a similar solution, but that might

                          Expires 10 June 1997          [Page 5]

Internet Draft    Internationalization of Domain Names  10 December 1996

   have other restrictions for the initial letters?]

   Please note that this solution has the following interesting proper-

   -  For subsequent positions, there is an equivalence between the hex-
      adecimal value of the character code and the target letter used.
      This assures easy conversion and checking.

   -  The absence of digits from the "initial" column, and the fact that
      the hyphen is not used, assures that the resulting string conforms
      to domain name syntax.

   -  Raw sorting of encoded and unencoded domain names is equivalent.

   -  The boundaries of characters can always be detected easily.
      (While this is important for representations that are used inter-
      nally for text editing, it is actually not very important here,
      because tools for editing can be assumed to use a more straight-
      forward representation internally.)

   -  Unless control characters are allowed, the target string will
      never actually contain a G.

3.3 Encoding Example

   As an example, the current domain


   with the components standing for information science, science, the
   University of Tokyo, academic, and Japan, might in future be repre-
   sented by


   (a transliteration of the kanji that might probably be chosen to rep-
   resent the same domain). Writing each character in U+HHHH notation as
   in [Unicode], this is


   and will be translated by the software handling internationalized
   domain names, according to the above specifications, to

                          Expires 10 June 1997          [Page 6]

Internet Draft    Internationalization of Domain Names  10 December 1996


3.4 Length Considerations

   DNS allows for a maximum of 63 positions in each part, and for 255
   positions for the overall domain name including dots.  This allows up
   to 15 ideographs, or up to 21 letters e.g.  from the Hebrew or Arabic
   alphabet, in a label.  While this does not allow for the same margin
   as in the case of ASCII domain names, it should still be quite suffi-
   cient.  [Problems could only surface for languages that use very long
   words or terms and don't know any kind of abbreviations or similar
   shortening devices. Do these exist?]  DNS contains a compression
   scheme that avoids sending the same trailing portion of a domain name
   twice in the same transmission. Long domain names are therefore not
   that much of a concern.

4. Usage Considerations

4.1 General Usage

   To implement this proposal, neither DNS servers nor resolvers need
   changes.  These programs will only deal with the encoded form of the
   domain name with the .i suffix. Software that wants to offer an
   internationalized user interface (for example a web browser) is
   responsible for the necessary conversions. It will analyze the domain
   name, call the resolver directly if the domain name conforms to the
   domain name syntax restrictions, and otherwise encode the name
   according to the specifications of Section 3.2 and append the .i suf-
   fix before calling the resolver.  New implementations of resolvers
   will of course offer a companion function to gethostbyname accepting
   a ISO10646/Unicode string as input.

4.2 Usage Restrictions

   While this proposal in theory allows to have control characters such
   as BEL or NUL or symbols such as arrows and smilies in domain names,
   such characters should clearly be excluded from domain names. Whether
   this has to be explicitly specified or whether the difficulty to type
   these characters on any keyboard of the world will limit their use

                          Expires 10 June 1997          [Page 7]

Internet Draft    Internationalization of Domain Names  10 December 1996

   has to be discussed.

   A related point is the question of equivalence. For historical rea-
   sons, ISO 10646/Unicode contain considerable number of compatibility
   characters and allow more than one representation for characters with
   diacritics. To guarantee smooth interoperability in these and related
   cases, additional restrictions or the definition of some form of nor-
   malization seem necessary. However, this is a general problem affect-
   ing all areas where ISO 10646/Unicode is used in identifiers, and
   should therefore be addressed in a generic way.

   Equally related is the problem of case equivalence.  Users can very
   well distinguish between upper case and lower case.  Also, casing in
   an i18n context is not as straightforward as for ASCII, so that case
   equivalence is best avoided.  Problems therefore result not from the
   fact that case is distinguished for i18n domain names, but from the
   fact that existing domain names do not distinguish case. Where it is
   impossible to distinguish between next.com and NeXT.com, the same two
   subdomains would easily be distinguishable if subordinate to a i18n

   A problem that also has to be discussed and solved is bidirectional-
   ity.  Arabic and Hebrew characters are written right-to-left, and the
   mixture with other characters results in a divergence between logical
   and graphical sequence. See [HTML-I18N] for more explanations.  The
   proposal of [Yer96] for dealing with bidirectionality in URLs could
   probably be applied to domain names.

4.3 Domain Name Creation

   The ".i" ZLD should be created as such to allow the internationaliza-
   tion of domain names. Rules for creating subdomains inside ".i"
   should follow the established rules for the creation of functionally
   equivalent domains in the existing domain hierarchy, and should
   evolve in parallel.  However, the peculiarities of i18n domain names
   should be carefully considered:

   -  Depending on the script, reasonable lengths for domain name parts
      may differ greatly. For ideographic scripts, a part may often be
      only a one-letter code. Established rules for lengths may need

   -  If the number of generic TLDs (.com, .edu, .org, .net) is kept
      low, then it may be feasible to restrict i18n TLDs to country

                          Expires 10 June 1997          [Page 8]

Internet Draft    Internationalization of Domain Names  10 December 1996

   -  There are no ISO 639 two-letter codes in scripts other than Latin.
      I18n domain names for countries will have to be designed from

   -  The names of some countries or regions may pose greater political
      problems when expressed in the native script than when expressed
      in 2-letter ISO 639 codes.

   -  I18n country domain names should in principle only be created in
      those scripts that are used locally. There is probably little use
      in creating an Arabic domain name for China, for example.

   -  In those cases where domain names are open to a wide range of
      applicants, a special procedure for accepting applications should
      be used so that a reasonable-quality fit between ASCII domain
      names and i18n domain names results where desired.  This would
      probably be done by establishing a period of about a month for
      applications inside a i18n domain newly created as a parallel for
      an existing domain, and resolving the detected conflicts.

   -  It may be desirable to have internationalized subdomains in non-
      internationalized TLDs. As an example, many companies in France
      may want to register an accented version of their company name,
      while remaining under the .fr TLD. For this, .fr would have to be
      reregistered as .M6N2.i. Accented and other internationalized sub-
      domains would go below .M6N2.i, whereas unaccented ones would go
      below .fr in its plain form.

   -  To generalize the above case, one might create a requirement that
      any domain name registry would be required to register and manage
      a corresponding .i domain upon request to allow registration of
      i18n domain names in arbitrary subdomains.

4.4 Usage in URLs

   According to current definitions, URLs encode sequences of octets
   into a sequence of characters from a character set that is almost as
   limited as the character set of domain names [RFC1738].  This is
   clearly not satisfying for i18n.

   Internationalizing URLs, i.e. assigning character semantics to the
   encoded octets, can either be done separately for each part and/or
   scheme, or in an uniform way. Doing it separately has the serious
   disadvantage that software providing user interfaces for URLs in gen-
   eral would have to know about all the different i18n solutions of the
   different parts and schemes. Many of these solutions may not even be

                          Expires 10 June 1997          [Page 9]

Internet Draft    Internationalization of Domain Names  10 December 1996

   known yet.

   It is therefore definitely more advantageous to decide on a single
   and consistent solution for URL internationalization. The most valu-
   able candidate [Yer96], for many reasons, is UTF-8 [RFC2044], an
   ASCII-compatible encoding of UCS4.

   Therefore, an URL containing the domain name of the example of Sec-
   tion 3.3 should not be written as:


   (although this will also work) but rather


   In this canonical form, the trailing .i is absent, and the octets can
   be reconstructed from the %HH-encoding and interpreted as UTF-8 by
   generic URL software. The software part dealing with domain names
   will carry out the conversion to the .i form.

5. Alternate Proposals

5.1 The Dillon Proposal

   The proposal of Michael Dillon [Dillon96] is also based on encoding
   Unicode into the limited character set of domain names. Distinction
   is done for each part, using the hyphen in initial position. Because
   this does not fully conform to the syntax of existing domain names,
   it is questionable whether it is backwards-compatible. On the other
   hand, this has the advantage that local i18n domain names can be
   installed easily without cooperation by the manager of the superdo-

   A variable-length scheme with base 36 is used that can encode up to
   1610 characters, absolutely insufficient for Chinese or Japanese.
   Characters assumed not to be used in i18n domain names are excluded,
   i.e. only one case is allowed for basic Latin characters.  This means
   that large tables have to be worked out carefully to convert between
   ISO 10646/Unicode and the actual number that is encoded with base 36.

                          Expires 10 June 1997         [Page 10]

Internet Draft    Internationalization of Domain Names  10 December 1996

5.2 Using a Separate Lookup Service

   Instead of using a special encoding and burdening DNS with i18n, one
   could build and use a separate lookup service for i18n domain names.
   Instead of converting to UCS4 and encoding according to Section 3.2,
   and then calling the DNS resolver, a program would contact this new
   service when seeing a domain name with characters outside the allowed

   Such a solution has various problems. A separate service does not yet
   exist, whereas DNS is readily usable. Solving the problems of unique-
   ness, etc., again for this separate service creates a lot of work. On
   the other side, there are no savings in terms of implementation
   costs. DNS also does not have a serious capacity problem that might
   be addressed by using a separate lookup service, nor is such a prob-
   lem created by i18n domain names.

6. Generic Considerations

6.1 Security Considerations

   This proposal is believed not to raise any other security considera-
   tions than the current use of the domain name system.

6.2 Internationalization Considerations

   This proposal addresses internationalization as such. The main addi-
   tional consideration with respect to internationalization may be the
   indication of language. However, for concise identifiers such as
   domain names, language tagging would be too much of a burden and
   would create complex dependencies with semantics.

        NOTE -- This section is introduced based on a recommenda-
        tion in [RFCIAB]. A similar section addressing internation-
        alization should be included in all application level
        internet drafts and RFCs.

                          Expires 10 June 1997         [Page 11]

Internet Draft    Internationalization of Domain Names  10 December 1996


   I am grateful in particular to the following persons:

   Bert Bos, Lori Brownell, Michael Dillon, David Goldsmith, Larry Mas-
   inter, Keith Moore, and Francois Yergeau


   [ASCII]        Coded Character Set -- 7-Bit American Standard Code
                  for Information Interchange, ANSI X3.4-1986.

   [Dillon96]     M. Dillon, "Multilingual Domain Names", Memra Software
                  Inc., November 1996 (circulated Dec. 6, 1996 on iahc-

   [HTML-I18N]    F. Yergeau, G. Nicol, G. Adams, and M. Duerst, "Inter-
                  nationalization of the Hypertext Markup Language",
                  Work in progress (draft-ietf-html-i18n-05.txt), August

   [ISO10646]     ISO/IEC 10646-1:1993. International standard -- Infor-
                  mation technology -- Universal multiple-octet coded
                  character Set (UCS) -- Part 1: Architecture and basic
                  multilingual plane.

   [RFC1034]      P. Mockapetris, "Domain Names - Concepts and Facili-
                  ties", ISI, Nov. 1987.

   [RFC1035]      P. Mockapetris, "Domain Names - Implementation and
                  Specification", ISI, Nov. 1987.

   [RFC1522]      K. Moore, "MIME (Multipurpose Internet Mail Exten-
                  sions) Part Two: Message Header Extensions for Non-
                  ASCII Text", University of Tennessee, September 1993.

   [RFC1642]      D. Goldsmith, M. Davis, "UTF-7: A Mail-safe Transfor-
                  mation Format of Unicode", Taligent Inc., July 1994.

   [RFC1738]      T. Berners-Lee, L. Masinter, and M. McCahill,
                   "Uniform Resource Locators (URL)", CERN, Dec. 1994.

   [RFC2044]      F. Yergeau, "UTF-8, A Transformation Format of Unicode
                  and ISO 10646", Alis Technologies, October 1996.

                          Expires 10 June 1997         [Page 12]

Internet Draft    Internationalization of Domain Names  10 December 1996

   [RFCIAB]       C. Weider, C. Preston, K. Simonsen, H. Alvestrand, R.
                  Atkinson, M. Crispin, P. Svanberg, "Report from the
                  IAB Character Set Workshop", October 1996 (currently
                  available as draft-weider-iab-char-wrkshop-00.txt).

   [Unicode]      The Unicode Consortium, "The Unicode Standard, Version
                  2.0", Addison-Wesley, Reading, MA, 1996.

   [Yer96]        F. Yergeau, "Internationalization of URLs", Alis Tech-

Author's Address

   Martin J. Duerst
   Department of Computer Science
   University of Zurich
   Winterthurerstrasse 190
   CH-8057 Zurich

   Tel: +41 1 257 43 16
   Fax: +41 1 363 00 35
   E-mail: mduerst@ifi.unizh.ch

     NOTE -- Please write the author's name with u-Umlaut wherever
     possible, e.g. in HTML as D&uuml;rst.

                                    Expires 10 June 1997         [Page 13]