This is only a rough draft - Megan 04/16/92 WAIS-W3-X.500 BOF MINUTES BOF at the March 1992 IETF[1] , on the evening of March 18. Summary This meeting followed discussion at the "living documents" BOF[2] the previous evening, and was more focussed in its discussion. The WAIS, World-Wide Web, Prospero systems for network information retrieval (NIR) were presented (the Gopher protocol was presented in plenary the following day). The x500 directory was presented in the light of NIR needs, as were two proposals to use the directory to refer to documents. A discussion followed as to how to allow these systems to inter-operate, and on requirements for name spaces. A working group was proposed to define the format for a generalized printable format for a name or address in any of these systems. Chair Steve Kille, UCL and ISODE consortium Present See list ietf-wwx-bof@info.cern.ch[3] . These minutes are available in hypertext form using WWW as http://info.cern.ch./hypertext/Conferences/IETF92/WWX_BOF_ mins.html as well as through the normal channels. WAIS John Curran of BBN presented the WAIS protocol, in the absence of anyone from Thinking Machines Corporation who were originally responsible for it. The WAIS model is of a number of servers, each of which serves a number of databases, each of which contains a number of documents. Client software allows many databases to be searched at the same time. The server keeps an inverted full text index for each database, so the search is very fast. Non-text files may also be served: recent extensions allow indexing of text files in new formats. The files indexed need not be copied, but the index is of the same order of size as the files. Many databases exist, but there is no scalable way of finding them (TMC currently keeps a master index). Use of x500 was discussed. The WAIS protocol is an extended subset of Z3950. The differences were discussed: WAIS allows relevance feedback ("Give me a document like this one") , and specifies how a query should be formulated. WAIS and Z39.50 have the same presentation layer. Documents in the Directory Wengiyk Yeongpresented his paper OSI-DS-22, "Representing public archives in the directory"[4]. His project puts information about documents, including the network address for retrieval, into the directory. He currently has RFCs and FYI documents in, but would like to move on to other internet archives. He concluded that he needed a more sophisticated approach. It was difficult to characterize arbitrary archives, with too little information about them. (See IAFA WG[5]). The World-Wide Web Tim Berners-Lee presented the World Wide Web (w3) and discussed requirements for interworking between the systems. The W3 project was initially funded to provide an information infrastructure to the world-wide community of high energy physicists. The data model is of documents which are hypertext and/or searchable indexes. The philosophy behind it is that a user should be able to point and click on phrase or a word within a document and the associated document would be retrieved from wherever in the world and presented to the user in an appropriate format - without the user having to be aware of where the document is located or what the access method is. These details are hidden in the hypertext links. There were server programs for many information servers, gateways to WAIS, Archie and gopher and client programs for various user machines. The W3 clients use several protocols for accessing documents (FTP, NNTP, WAIS, Gopher, and W3's own "HTTP") although this is hidden from the user. The HTTP protocol is a simple stateless search/retrieve protocol running over TCP. As originally conceived but not yet implemented, it included authentication and data format negotiation.Tim discussed the differences between WWW, WAIS, Archie, Gopher and Prospero systems. The need for a Universal Document Identifier (UDI) for describing the address or, given a directory, name, for a document whatever is access protocol was discussed, as outlined in OSI-DS-XX. Each application uses a "handle" for a file which can be prefixed by the particular protocol name to generate a universal address. Most systems (WAIS excepted) are extensible, entertaining document addresses which refer to other systems. WAIS indexes currently can only refer to documents in the same database, let alone with other retrieval methods. There is a need for WAIS to be more flexible. John Curran said he would bring this to the attention of the WAIS community. Addresses would not in the long term be suitable for references to documents, so it was hoped that some sort of directory service, operating within the UDI framework, would be incorporated. More information: telnet info.cern.ch. Client and server code is available by anonymous FTP from info.cern.ch. Mailing lists: www-talk@info.cern.ch, www-interest@info.cern.ch Discussion document: OSI-DS-29[6] Representing the Real World in the Directory Paper: OSI-DS-25[7]Steve Kille discussed this paper "Representing the Real World in an X.500 Directory". A Listing Service may be used to group like information items together for example to provide a Yellow Pages Service. Such a service could for example provide for members of a special interest group, or could group documents on a particular subject.Services such as Archie could be considered to be Listing Services. One imagines an information Universe in which Information Brokers provide different subject based (say) views via their listing service. One would then need to locate the various listing services (using a mechanism such as a directory?) UK British Library Project Paul Barker described a project, sponsored by the British Library, to represent grey literature (unpublished research papers) in the Directory. The project is thought to be unlikely to succeed - but one of the aims is to demonstrate whether or not it is possible. They will take the (UK) MARC records and model these within X.500. They might also consider trying to provide a listing service so that the documents might be retrieved more readily by subject area. Prospero Cliff Neuman described Prospero. It follows a file system model, rather than the hypertext model. It is built on UDP for speed. It has the notion of a Directory which contains links to other objects (other directories or files). It returns the link to the information object and then automatically retrieves the file by another mechanism by the appropriate access method (Archie, WAIS, nntp, WWW - soon!, NFS, ftp etc.) It has been used very successfully to access the archie database. Cliff stated that he expected to be able to use X.500 to translate between the document ID and how to get the document. With Prospero the user has his own view of the global information base (or has a view built for him). Cliff thought there should be multiple name spaces - but the difficulty would be that these would need representing near the top of the directory tree. With multiple user chosen views - this would be difficult to manage. Also two users might refer to an object by different handles which would be relative to their individual name spaces - difficult when passing references (say in a mail message) from one person to the other. The concept of "Closure": Each object has a related name space. All references within the object are resolved using the context of the name space. Name spaces themselves have global network addresses, but the user doesn't see that. More information: info-prospero@isi.edu System 33 Larry Masinter talked about a project at Xerox PARC. This has the concepts: HANDLE 32 byte number (is a content ID). In fact this contains hints for finding the document. FILE Location (6 part) Protocol; Host; Path; piece; format; timeout Description (normal "Catalogue" information: Name, Author, etc) There is format negotiation when a document is retrieved. It is not simple in reality to categorize data formats as there is such a plethora of different varieties. Gateways provide access between systems not sharing transport protocols. Also considered Access Control. ACL is part of description. The Server exploits multiple protocols for Search and retrieve. There is a problem with dealing with different types of document (applications for jobs, product specs, memos, contracts, faxes, etc. ) It is difficult to normalize the attributes of a general document. Summing up Tim Berners-Lee summed up by saying that all applications described used resolvable document address, and so for interworking, we need a universal representation for such a network object address. With the coming of directories, names should increasingly be used in place of network addresses. The Universal Document Identifier was intended to be able to hold either an name or address for any access protocol. (This is not the same as "USDN" a document serial number which is not resolvable, but only one of which exists for each document). In discussion, Steve Kille suggested should be a WG on details of UDIs and a separate one for USDN. A comment was that the W3 data model encompasses those of the other systems. John Curran insisted on a better term than "UDI", suggesting "Document Access Token". Peter Deutch's need for a USDN is to be able to determine the equivalence of two USDN. Chris Weider agreed to co-author a document on the issues. Jill Foster suggested a pilotproject to put UDI's in the directory for a set of documents and to have the gopher, Prospero, archie, and Prospero people try to utilise these. [These minutes have been largely built from Jill Foster's report[8] and Karen Sollins' notes[9] for which I am most grateful, though errors in the above are probably mine. Tim BL] References: [1]http://info.cern.ch/hypertext/Conferences/IETF92/IETF-9203.html [2]http://info.cern.ch/hypertext/Conferences/IETF92/LivingDocuments.h tml [3]http://info.cern.ch/hypertext/WWW/Administration/Mailing/ietf-wwx- bof [4]file://cs.ucl.ac.uk/osi-ds/osi-ds-22-00.txt [5]http://info.cern.ch/hypertext/Conferences/IETF92/IAFA-BOF.html [6]file://cs.ucl.ac.uk/osi-ds/osi-ds-29-00.txt [7]file://cs.ucl.ac.uk/osi-ds/osi-ds-25-00.txt [8]http://info.cern.ch/hypertext/Conferences/IETF92/WWX_BOF.html [9]http://info.cern.ch/hypertext/Conferences/IETF92/WWX_BOF_Sollins.h tml