Rfc1727
TitleA Vision of an Integrated Internet Information Service
AuthorC. Weider, P. Deutsch
DateDecember 1994
Format:TXT, HTML
Status:INFORMATIONAL






Network Working Group                                          C. Weider
Request for Comments: 1727                                    P. Deutsch
Category: Informational                       Bunyip Information Systems
                                                           December 1994


         A Vision of an Integrated Internet Information Service

Status of this Memo

   This memo provides information for the Internet community.  This memo
   does not specify an Internet standard of any kind.  Distribution of
   this memo is unlimited.

Abstract

   This paper lays out a vision of how Internet information services
   might be integrated over the next few years, and discusses in some
   detail what steps will be needed to achieve this integration.

Acknowledgments

   Thanks to the whole gang of information service wonks who have
   wrangled with us about the future of information services in
   countless bar bofs (in no particular order): Cliff Lynch, Cliff
   Neuman, Alan Emtage, Jim Fullton, Joan Gargano, Mike Schwartz, John
   Kunze, Janet Vratny, Mark McCahill, Tim Berners-Lee, John Curran,
   Jill Foster, and many others. Extra special thanks to George Brett of
   CNIDR and Anders Gillner of RARE, who have given us the opportunity
   to start tying together the networking community and the librarian
   community.

1. Disclaimer

   This paper represents only the opinions of its authors; it is not an
   official policy statement of the IIIR Working Group of the IETF, and
   does not represent an official consensus.

2. Introduction

   The current landscape in information tools is much the same as the
   landscape in communications networks in the early 1980's.  In the
   early 80's, there were a number of proprietary networking protocols
   that connected large but autonomous regions of computers, and it was
   difficult to coalesce these regions into a unified network. Today, we
   have a number of large but autonomous regions of networked
   information.  We have a vast set of FTPable files, a budding WAIS
   network, a budding GOPHER network, a budding World Wide Web network,



RFC 1727                 Resource Transponders             December 1994


   etc.  Although there are a number of gateways between various
   protocols, and information service providers are starting to use
   GOPHER to provide a glue between various services, we are not yet in
   that golden age when all human information is at our fingertips. (And
   we're even farther from that platinum age when the computer knows
   what we're looking for and retrieves it before we even touch the
   keyboard.)

   In this paper, we'll propose one possible vision of the information
   services landscape of the near future, and lay out a plan to get us
   there from here.

3. Axioms of information services

   There are a number of unspoken assumptions that we've used in our
   discussions.  It might be useful to lay them out explicitly before we
   start our exploration.

   The first is that there is no unique information protocol that will
   provide the flexibility, scale, responsiveness, worldview, and mix of
   services that every information consumer wants.  A protocol designed
   to give quick and meaningful access to a collection of stock prices
   might look functionally very different from one which will search
   digitized music for a particular musical phrase and deliver it to
   your workstation. So, rather than design the information protocol to
   end all information protocols, we will always need to integrate new
   search engines, new clients, and new delivery paradigms into our
   grand information service.

   The second is that distributed systems are a better solution to
   large-scale information systems than centralized systems.  If one
   million people are publishing electronic papers to the net, should
   they all have to log on to a single machine to modify the central
   archives? What kind of bandwidth would be required to that central
   machine to serve a billion papers a day?  If we replicate the central
   archives, what sort of maintenance problems would be encountered?
   These questions and a host of others make it seem more profitable at
   the moment to investigate distributed systems.

   The third is that users don't want to be bothered with the details of
   the underlying protocols used to provide a given service. Just as
   most people don't care whether their e-mail message gets split up
   into 20 packets and routed through Tokyo to get to its destination,
   information service users don't care whether the GOPHER server used
   telnet to get to a WAIS database back-ended by an SQL database.  They
   just want the information. In short, they care very much about how
   they interact with the client; they just don't want to know what goes
   on behind.



RFC 1727                 Resource Transponders             December 1994


   These axioms force us to look at solutions which are distributed,
   support multiple access paradigms, and allow information about
   resources to be handed off from one system (say Gopher) to another
   (say WWW).

4. An architecture to provide interoperability and integration.

   The basic architecture outlined in this paper splits up into 4 levels
   [Fig. 1].

   At the lowest level, we have the resources themselves. These are such
   things as files, telnet sessions, online library catalogs, etc. Each
   resource can have a resource transponder attached [Weider 94a], and
   should have a Uniform Resource Name (URN) [Weider 94b] associated
   with it to uniquely identify its contents. If a resource transponder
   is attached, it will help maintain the information required by the
   next level up.

   At the next level, we have a 'directory service' that takes a URN and
   returns Uniform Resource Locators (URLs) [Berners-Lee 94] for that
   resource. The URL is a string which contains location information,
   and can be used by a client to access the resource pointed to by the
   URL.  It is expected that a given resource may be replicated many
   times across the net, and thus the client would get a number of URLs
   for a given resource, and could choose between them based on some
   other criteria.

























RFC 1727                 Resource Transponders             December 1994


     ______________________________________________________________
     |           |              |       |               |
     |           |              |       |               |
     |  Gopher   |  WAIS        | WWW   | Archie        | Others ...
     |           |              |       |               |
     |___________|______________|_______|_______________|___________
          |                                |
          |                       _________|____________
          |                      |                      |
          |                      | Resource Discovery   |
          |                      |  System (perhaps     |
          |                      |  based on whois++)   |
          |                      |______________________|
          |                                |
          |                                |
     _____|________________________________|____
    |                                           |
    | Uniform resource name to uniform resource |
    | locator mapping system (perhaps based on  |
    | whois++ or X.500)                         |
    |___________________________________________|
                        |
                        |
        ________________|______________________________________
        |                  |                 |                 |
  ______|______     _______|_____      ______|______     ______|______
 |             |   |             |    |             |   |             |
 | Transponder |   | Transponder |    | Transponder |   | Transponder |
 |_____________|   |_____________|    |_____________|   |_____________|
 |             |   |             |    |             |   |             |
 |             |   |             |    |             |   |             |
 |             |   |             |    |             |   |             |
 |  Resource   |   |  Resource   |    |  Resource   |   |  Resource   |
 |             |   |             |    |             |   |             |
 |             |   |             |    |             |   |             |
 |_____________|   |_____________|    |_____________|   |_____________|


        Figure 1: Proposed architecture of an integrated information
                        service

   The third level of the architecture is a resource discovery system.
   This would be a large, distributed system which would accept search
   criteria and return URNs and associated information for every
   resource which matched the criteria. This would provide a set of URLs
   which the information service providers (GOPHER servers, etc.) could
   then select among for incorporation.




RFC 1727                 Resource Transponders             December 1994


   The fourth level of the architecture is comprised of the various
   information delivery tools.  These tools are responsible for
   collating pointers to resources, informing the user about the
   resources to which they contain pointers, and retrieving the
   resources when the user wishes.

   Let's take a look in greater detail at each of these levels.

4.1 Resource layer

   The resources at this layer can be any collection of data a publisher
   wishes to catalog. It might be an individual text file, a WAIS
   database, the starting point for a hypertext web, or anything else.
   Each resource is assigned a URN by the publisher, and the URL is
   derived from the current location of the resource. The transponder is
   responsible for updating levels 2 and 3 with the appropriate
   information as the resource is published and moves around.

4.2 URN -> URL mapping

   This level takes a URN and returns a number of URLs for the various
   instantiations of that resource on the net.  It will also maintain
   the URN space. Thus the only functionality required of this level is
   the ability to maintain a global namespace and to provide mappings
   from that namespace to the URLs. Consequently, any of the distributed
   'directory service' protocols would allow us to provide that service.
   However, there may be some benefit to collapsing levels 2 and 3 onto
   the same software, in which case we may need to select the underlying
   protocol more carefully. For example, X.500 provides exactly the
   functionality required by level 2, but does not (yet) have the
   functionality required to provide the level 3 service.  In addition,
   the service at level 2 does not necessarily have to be provided by a
   monolithic system. It can be provided by any collection of protocols
   which can jointly satisfy the requirements and also interoperate, so
   that level 2 does appear to level 3 to be universal in scope.

4.3 Resource discovery

   This is the level which requires the most work, and where the
   greatest traps lurk to entangle the unwary. This level needs to serve
   as a giant repository of all information about every publication,
   except for that which is required for the URI -> URL mapping. Since
   this part is the least filled in at the moment, we will propose a
   mechanism which may or may not be the one which is eventually used.

   When a new resource is created on the network, it is assigned a URN
   determined by the publisher of the resource. Section 4.1 discusses in
   more detail the role of the publisher on the net, but at the moment



RFC 1727                 Resource Transponders             December 1994


   we can consider only 2 of the publisher's functions. The publisher is
   responsible for assigning a URN out of the publishers namespace, and
   is responsible for notifying a publishing agent [Deutsch 92] that a
   new resource has been created; that agent will either be a part of
   the resource location service or will then take the responsibility
   for notifying an external resource location service that the resource
   has been created. Alternatively, the agent can use the resource
   location service to find parts of the RLS which should be notified
   that this resource has been created.

   To give a concrete example, let's say that Peter and Chris publish a
   multi- media document titled, "Chris and Peter's Bogus Journey",
   which talks about our recent trip to the Antarctic, complete with
   video clips. P & C would then ask their publishing agent to generate
   a URN for this document. They then ask their publishing agent to
   attach a transponder to the document, and to look around and see if
   anyone a) has asked that our agent notify them whenever anything we
   write comes out; or b) is running any kind of server of 'trips to
   Antarctica'. Janet has posted a request that she be notified, so the
   agent tells her that a new resource has been created. The agent also
   finds 3 servers which archive video clips of Antarctica, so the agent
   notifies all three that a new resource on Antarctica has come out,
   and gives out the URN and a URL for the local copy.

4.4 Information delivery tools

   One of the primary functions of an information delivery tool is to
   collect and collate pointers to resources. A given tool may provide
   mechanisms to group those pointers based on other information about
   the resource, e.g.  a full-text index allows one to group pointers to
   resources based on their contents; archie can group pointers based on
   filenames, etc. The URLs which are being standardized in the IETF are
   directly based on the way World Wide Web built pointers to resources,
   by creating a uniform way to specify access information and location
   for a resource on the net. With just the URLs, however, it is
   impossible without much more extensive checking to tell whether two
   resources with different URLs have the same intellectual content or
   not. Consequently, the URN is designed to solve this problem.

   In this architecture, the pointers that a given information delivery
   tool would keep to a resource will be a URN and one or more cached
   URLs. When a pointer to a resource is first required (i.e. when a new
   resource is linked in a Gopher server), level 2 will provide a set of
   URLs for that URN, and the creator of the tool can then select which
   of those will be used. As it is expected that the URLs will
   eventually become stale (the resource moves, the machine goes down,
   etc.) the URN can be used to get a set of current URLs for the
   resource and an appropriate one can then be selected. Since the cost



RFC 1727                 Resource Transponders             December 1994


   of using the level 2 service is probably greater than the cost of
   simply resolving a URL, both the URN and the URL are cached to
   provide speedy access unless something has moved.

4.5 Using the architecture to provide interoperability between services

   In the simplest sense, each interaction that we have with an
   information delivery service does one of two things: it either causes
   a pointer to be resolved (a file to be retrieved, a telnet session to
   be initiated, etc.) or causes some set of the pointers available in
   the information service to be selected. At this level, the
   architecture outlined above provides the core implementation of
   interoperability. Once we have a means of mapping between names and
   pointers, and we have a standard method of specifying names and
   pointers, the interoperation problem becomes one of simply handing
   names and pointers around between systems. Obviously with such a
   simplistic interoperability scheme much of the flavor and
   functionality of the various systems are lost in transition. But,
   given the pointers, a system can either a) present them to the user
   with no additional functionality or b) resolve the pointers, examine
   the resources, and then run algorithms designed to tie these
   resources together into a structure appropriate for the current
   service. Let's look at one example (which just happens to be the
   easiest to resolve); interoperation between World Wide Web and
   Gopher.

   Displaying a Gopher screen as a WWW document is trivial with these
   pointers.  Every Gopher screen is simply a list of menu items with
   pointers behind them (we'll ignore the other functionality Gopher
   provides for a moment), so is an extremely simple form of a hypertext
   document. Consequently with this architecture it is easy to show and
   resolve a Gopher screen in WWW.  For a WWW to Gopher map, the
   simplest method would be that when one accesses a WWW document, all
   the pointers associated with links off to other documents are brought
   in with the document. Gopher could then resolve the links and read
   the first line of each document to provide a Gopher style screen
   which contains everything in the WWW document. When a link is
   selected, all of the WWW links for the new document are brought in
   and the process repeats. Obviously we're losing a lot with the WWW ->
   Gopher mapping; some might argue that we are losing everything.
   However, this does provide a trivial interoperability capacity, and
   one can argue that the 'information content' has been preserved
   across the gateway.

   In addition, the whole purpose of gatewaying is to provide access to
   resources that lie outside the reach of your current client. Since
   all resources are identifiable and accessible through layers 2 and 3,
   it will be easy to copy resources from one protocol to another since



RFC 1727                 Resource Transponders             December 1994


   all we need to do is to move the pointers and reexpress the
   relationships between the pointers in the new paradigm.

4.6 Other techniques for interoperability

   One technique for interoperability which has recently received some
   serious attention is the technique of creating one client which
   speaks the protocols of all the information delivery tools. This
   approach has been taken in particular by the UNITE (User's Network
   Interface To Everything) group in Europe. This client would sit on
   the top level of the architecture in Figure 1. This technique is best
   exemplified by the recent work which has gone into Mosaic, a client
   which can speak almost all of the major information services
   protocols. This technique has a lot of appeal and has enjoyed quite a
   bit of success; however, there are several practical difficulties
   with this approach which may hinder its successful implementation.

   The first difficulty is one that is common to clients in general; the
   clients must be constantly updated to reflect changes in the
   underlying protocols and to accommodate new protocols. If the
   increase in the number of information services is very gradual, or if
   the underlying protocols do not change very rapidly, this may not be
   an insuperable difficulty. In addition, old clients must have some
   way of notifying their user that they are no longer current;
   otherwise they will no longer be able to access parts of the
   information mesh.

   The second problem is one which may prove more difficult. Each of the
   currently deployed information services provides information in a
   fundamentally different way. In addition, new information services
   are likely to use completely new paradigms for the organization and
   display of the information they provide. The various clients of these
   information services provide vastly different functionality from each
   other because the underlying protocols allow different techniques. It
   may very well prove impossible to create a single client which allows
   access to the full functionality of each of the underlying protocols
   while presenting a consistent user interface to the user.

   Much of the success of Mosaic and other UNITE tools is due to the
   fact that Gopher, WWW, and other tools are still primarily text
   based. When new tools are deployed which depend more on visual cues
   than textual cues, it may be substantially more difficult to
   integrate all these services into a single client.

   We will continue to follow this work and may include it in future
   revisions of this architecture if it bears fruit.





RFC 1727                 Resource Transponders             December 1994


5. Human interactions with the architecture

   In this section we will look at how humans might interact with an
   information system based on the above architecture.

5.1 Publishing in this architecture

   When we speak of publishing in this section, we are referring only to
   the limited process of creating a resource on the net, assigning it a
   URN, and spreading the information around that we have created a new
   resource.

   We start with the creation of a resource. Simple enough; a creative
   thought, a couple of hours typing, and a few cups of coffee and we
   have a new resource.  We then wish to assign it a URN. We can talk to
   whichever publishing agent we would like; whether it is our own
   personal publishing agent or some big organization that provides URN
   and announcement services to a large number of authors.  Once we have
   a URN, we can provide the publishing agent with a URL for our local
   copy of the resource and then let it do its thing.  Alternatively, we
   can attach a transponder to the resource, let it determine a local
   URL for the resource, and let it contact the publishing agent and set
   up the announcement process. One would expect a publishing agent to
   prompt us for some information as to where it should announce our new
   resource.

   For example, we may just wish a local announcement, so that only
   people in our company can get a pointer to the resource. Or we may
   wish some sort of global announcement (but it will probably cost us a
   bit of money!)

   Once the announcement has been made, the publishing agent will be
   contacted by a number of pieces of the resource location system. For
   example, someone running a WAIS server may decide to add the resource
   to their index. So they can retrieve the resource, index it, and add
   the indexes to their tables along with a URI - URL combination. Then
   when someone uses that WAIS server, it can go off and retrieve the
   resource if necessary. Or, the WAIS server could create a local copy
   of the resource; if it wished other people to find their local copy
   of the resource, it could provide the URI -> URL mapper with a URL
   for the local copy. In any case, publication becomes a simple matter.

   So, where does this leave the traditional publisher? Well, there are
   a number of other functions which the traditional publisher provides
   in addition to distribution. There are editorial services, layout and
   design, copyright negotiations, marketing, etc.  The only part of the
   traditional role that this system changes is that of distributing the
   resource; this architecture may make it much cheaper for publishers



RFC 1727                 Resource Transponders             December 1994


   to distribute their wares to a much wider audience.

   Although copying of resources would be possible just as it is in
   paper format, it might be easier to detect republication of the
   resource in this system, and if most people use the original URN for
   the resource, there may be a reduced monetary risk to the publisher.

5.2 A librarian role in this architecture

   We've been in a number of discussions with librarians over the past
   year, and one question that we're frequently asked is "Does Peter
   talk this rapidly all the time?". The answer to that question is
   "Yes". But another question we are frequently asked is "If all these
   electronic resources are going to be created, supplanting books and
   journals, what's left for the librarians?".  The answer to that is
   slightly more complex, but just as straightforward.  Librarians have
   excelled at obtaining resources, classifying them so that users can
   find them, weeding out resources that don't serve their communities,
   and helping users navigate the library itself. None of these roles
   are supplanted by this architecture. The only differences are that
   instead of scanning publisher's announcements for new resources their
   users might be interested in, they will have to scan the
   announcements on the net. Once they see something interesting, they
   can retrieve it (perhaps buying a copy just as they do now), classify
   it, set up a navigation system for their classification scheme, show
   users how to use it, and provide pointers (or actual copies) of the
   resource to their users. The classification and selection processes
   in particular are services which will be badly needed on a net with a
   million new 'publications' a day, and many will be willing to pay for
   this highly value added service.

5.3 Serving the users

   This architecture allows users to see the vast collection of
   networked resources in ways both familiar and unfamiliar. Bookstores,
   record shops, and libraries can all be constructed on top of this
   architecture, with a number of different access methods. Specialty
   shops and research libraries can be easily built, and then tailored
   to a thousand different users.  One never need worry that a book has
   been checked out, that a CD is out of stock, that a copy of Xenophon
   in the original Greek isn't available locally.  In fact, a user could
   even engage a proxy server to translate resources into forms that her
   machine can use, for example to get a text version of a Postscript
   document although her local machine has no Postscript viewer, or to
   obtain a translation of a sociology paper written in Chinese.

   In any case, however the system looks in three, five, or fifty years,
   we believe that the vision we've laid out above has the flexibility



RFC 1727                 Resource Transponders             December 1994


   and functionality to start tying everything together without forcing
   everyone to use the same access methods or to look at the net the
   same way. It allows new views to evolve, new resources to be created
   out of old, and for people to move from today to tomorrow with all
   the comforts of home but all the excitement of exploring a new world.

6. References

   [Berners-Lee 93] Berners-Lee, T., Masinter, L., and M. McCahill,
   Editors, "Universal Resource Locators", RFC 1738, CERN, The Xerox
   Corporation, University of Minnesota, December 1994.

   Deutsch, P., Master's Thesis, June 1992.
   Available for anonymous FTP as
   <ftp://archives.cc.mcgill.ca/pub/peterd/peterd.thesis>.

   [Weider 94a] Weider, C., "Resource Transponders", RFC 1728, Bunyip
   Information Systems, December 1994.

   [Weider 94b] Weider, C. and P. Deutsch, "Uniform Resource Names",
   Work in Progress.

Security Considerations

   Security issues are not discussed in this memo.

7. Authors' Addresses

   Chris Weider
   Bunyip Information Systems, Inc.
   2001 S. Huron Parkway #12
   Ann Arbor, MI 48104

   Phone: +1 313-971-2223
   EMail: clw@bunyip.com


   Peter Deutsch
   Bunyip Information Systems, Inc.
   310 Ste. Catherine St. West, Suite 202
   Montreal, Quebec, CANADA

   Phone: +1 514-875-8611
   EMail: peterd@bunyip.com