Integration with Document Digital Libraries

The Corporation for National Research Initiatives (CNRI) is working with five major universities (CMU, Cornell, UC-Berkeley, Stanford, and MIT) on an ARPA-sponsored project to develop concepts for digital libraries. As part of this project, each university is placing its Computer Science Technical Reports on-line and providing access to the distributed CSTR collection. Technologies developed for the CSTR project include the Dienst distributed search system [18] and a Handle Management Service for assigning, maintaining, and using unique identifiers for digital library objects [16].

The basic architecture being developed by CNRI for distributed digital libraries includes the following concepts [26]:

• A digital object which consists of a sequence of bits plus a unique identifier known as a handle (the binding between the handle and the sequence of bits may change over time).
• Naming authorities who are responsible for assigning unique identifiers within their portions of the handle namespace.
• Repositories from which digital objects are available.
• Information and Reference (IR) servers that provide reference information about collections of digital objects.

The CNRI work is closely related to the IETF Uniform Resource Identifier (URI) Working Group's work on Uniform Resource Names (URNs) [32] and Uniform Resource Citations [17]. CNRI's handle is the equivalent of IETF's URN, and CNRI's IR server serves a similar function to IETF's URC server.

The Netlib and NHSE Development Group has been engaged in a parallel effort to implement a location-independent naming architecture [11]. We provide for two types of location-independent names:

• a Uniform Resource Name (URN), for which the contents it refers to may change - e.g., the ``current version of LAPACK''.
• a Location Independent File Name (LIFN), for which the binding between the name and the byte contents of the file it refers to is fixed, once assigned. This type of name is needed for unambiguous references when attaching critical reviews or reporting scientific results obtained using a particular version of a piece of software. LIFNs also permit reliable and efficient cacheing and mirroring of files.

We are also developing a URC server that provides support for the following:

• Provision by the publisher of attribute-value pairs for a given URN in the form of cryptographically signed assertions.
• Retrieval and authentication of assertions by users.
• Specification of the data model used for a particular URN.
• Choice of encryption algorithm, including none.

We propose to integrate our software repository naming architecture with CNRI's digital library architecture in the following manner:

1. Both URNs and LIFNs will be expressible as handles, and URN and LIFN lookup will be merged with the Handle Management Service.
2. Our URC server will be an implementation of CNRI's IR server that may be used for cataloguing general Web resources, including software and data archives.
3. Similar to the Dienst protocol for document repositories, we will develop service specifications and retrieval protocols appropriate for software and data repositories. In addition, similar to the Digital Library Document Architecture that defines requirements for digital document structure [33], we will define requirements for software and data archive structures.