Network Working Group
Request for Comments: 1728
Category: Informational
C. Weider
Bunyip Information Systems
December 1994

Resource Transponders

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.


Although a number of systems have been created in the last several years to provide resource location and navigation on the Internet, the information contained in these systems must be maintained and updated by hand. This paper describes an automatic mechanism, the resource transponder, for maintaining resource location information.

Author's Note:

This document is being circulated as sort of a research paper; consequently there are no protocol specifications or anything of the sort. I hope that we can go from here and actually design them if there's consensus that they are potentially useful. Once we have some idea of the required functionality, we can then go out and standardize them.


This paper represents only the opinions of the author; it does not represent the consensus of the IIIR Working Group, although it is recognized by them as one legitimate approach to a solution of the problem.

1. Introduction

In the past few years, we've seen the invention and growth of a number of information location systems on the Internet, e.g., archie, Gopher, and WAIS. However, as these systems have become widely deployed, a number of maintenance and security problems have arisen with them. Some of the major ones:

  1. Out of necessity, most of these systems contain pointers to the desired resources rather than the resources themselves. Therefore, if a resource becomes obsolete, is modified, or is moved, the location system must be updated by hand. Some systems (archie in particular) proactively create updated indexes by contacting every resource on a certain time schedule (every 30 days or so) but this means that the system can be up to 30 days out of date, and this process can be highly inefficient depending on the percentage of information that has changed.
  1. Conversely, anyone who maintains a resource that they wish indexed must keep track of every directory which contains a pointer to that resource, so that if it is modified, all the directories can be updated. This obviously is an optimistic scenario.
  1. Many organizations which have installed these systems do not have the the available resources or expertise to maintain the information in the systems. Thus we have long periods where the information drifts, then a short period when the information is updated again.
  1. Even though these systems are almost always out of date today, this problem will become increasingly harder for humans to manage by hand as everyone on the net becomes their own publisher. Also, as the net speeds up and people rely more and more on accurate information, human-induced delays in updates of these systems will become increasingly intolerable.
  1. Most, if not all, of these systems provide no security whatsoever; if a pointer to a resource appears in a locator system, then it is assumed to be meant for public consumption. There are many potential information providers who would like to use publicly deployed information systems to publish to a very selected clientele, and do not wish to allow the whole net access to their resources.

2. Requirements for a Solution

There are several objectives which must be met by any proposed solution to these problems:

  1. We need to decrease the personnel resources needed for indexing and pointer maintenance.
  1. We need to increase the reliability and accuracy of the information held in resource location systems.
  1. We need to provide some mechanisms for security, particularly by mediating access to the resources.
  1. We need to make it easy for non-experts, such as librarians, archivists, and database maintainers, to announce their new resources to the various resource location services.

Many of these problems can be solved by a 'resource transponder' mechanism.

3. Resource Transponders

The resource transponder system works by adding two new layers to every resource: metainformation and an agent to update a resource location system (RLS) with that metainformation. The metainformation layer is physically attached to every resource, so that when the resource is moved or altered, the metainformation is immediately available to update the RLS. The agent layer may also be attached to the resource or may not be; the implications of both of these options are discussed in detail below.

   3.1 Metainformation

The metainformation layer of a given resource contains any information which might be required to create a pointer to this resource, and any information which may be useful for indicating how to catalog or index the resource. For example, the metainformation layer of a text document might contain such things as the Uniform Resource Name (URN) of the document (this is sort of a ISBN number for electronic resources), the title of the document, a Uniform Resource Locator (URL) for the document (this is a combination net address and access method indicator, used for retrieval), the size of the document, etc. Thus the metainformation layer contains data about the resource to which it is attached.

This metainformation is expected to be modifiable. For example, the metainformation layer may contain a history of where this particular copy of a resource has been. Let's say that a resource/transponder pair has been moved. When it gets to its new location, the agent can then attempt to contact the resource at its old location to determine whether the resource is still there (in which case the agent will simply cause the new location to be added to the RLS) or whether the resource is not there (in which case the agent can tell the RLS to add the current pointer and delete the old one).

A number of other possibilities for the contents of the metainformation level are contained in section 4.1.

3.2 Agents

The agent layer of a given resource contains an executable program which is responsible for reading the metainformation attached to the resource and using that information to update a RLS. It is also responsible for updating the metainformation where necessary and for running any indexing programs required by the RLS it is attempting to update.

When the tools required to build agents are constructed and deployed, the author expects the agents to begin mediating access to the resource, particularly for agents attached to resources which are not currently considered active processes, such as text files and digitized images. In this futuristic model, someone wishing to read a given document would have to first negotiate access to the data with the agent; the agent would then be responsible for delivering the data to the client. However, it is expected that this type of agent will not be widely deployed for some time.

Different ways of implementing agents are discussed in section 4.2.

4. Models for implementations of resource transponders

4.1. Models for implementations of the metainformation layer

The metainformation layer can be impelemented in a number of ways, depending on the resource with which it is associated. For an 'active' resource, such as an on-line catalog or a mail-based service, the metainformation can be stored in a file with a well- known name in the software distribution. Alternatively, the metainformation could be stored as a record in the data which the resource serves. For a text document, the metainformation could be stored as the first or last N bytes of the document (which would break a number of editors and file display techniques, but would guarantee that the metainformation is moved with the resource), or perhaps as a file with a logically associated name (paper2.meta associated with paper2.txt, for example). The problem with this second approach is that the user must know that they have to move the metainformation with the file itself, or things will start breaking. If an agent is explicitly attached to the resource, the agent could contain the metainformation internally.

In any case, the resource transponder system must be able to guarantee that the metainformation is moved when the resource is moved.

4.2 Models for implementations of the agents

The agent layer can also be implemented in a number of ways, depending on such things as system loads, desired sizes of resources, multitasking capabilities, etc.

The easiest and for many unitasking systems the cleanest way of implementing an agent is to have one agent per computer. Then when a resource is moved onto that computer, the agent is explicitly activated and notified where the new resource is. For example, let's say that someone wishes to download a copy of a resource and then let the RLS know that that resource is available for public consumption. She would download the resource and then run the agent, which would then notify the RLS and update the metainformation attached to the resource. This model could also be used to track files on a LAN, or to provide local location services with no need to run a larger RLS.

Another model for implementation of the agent is to have one agent per resource. In this model, the agent would be moved along with the resource and the metainformation. The agent could be implemented in a file which would be associated with the resource; in that case the agent would have to be explicitly activated when the resource was moved. Alternatively, the agent/metainformation/resource system could be implemented as one system, or in one file. In this case, the agent itself would always be active, and would be responsible for mediating access to the resource. When one did a 'telnet' to a resource with an active agent, the agent would accept the telnet connection and be responsible for providing security and translation for the data. This could provide great security for resources while still allowing pointers to them to be placed in public RLS's; the data in the resource could be encrypted, with the agent responsible for decrypting it.

5. Security Considerations

Security issues are discussed throughout this memo.

6. Author's Address

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

   Phone: +1 313-971-2223
   Fax: +1 313-971-2223