CNI Testbed Meeting, Omni Georgetown, Washington, DC February 10,1992 1. Introductions Attendees: Clifford Lynch, University of California (Chair) Cedric Bennett, Stanford University Hans Breitenlowner University of Maryland George Brett, MCNC/University of North Carolina Bill Cattey, MIT Information Systems Ray Denenberg, Library of Congress Larry Dixson, Library of Congress Sean Donelan, Data Research Eric Ferrin, Penn State John Kunze, University of California Berkeley Ron Larsen, University of Maryland James Michael, Data Research Harry Morris, Thinking Machines Mark Needleman, University of California Ralph Orlik, Library of Congress Bo Parker, Stanford University Cecilia Preston., University of California Berkeley (Notes) Sara Randall, Notis Systems Daivd Richards, Research Libraries Group Peter Ryall, Mead Data Central Bob Shafer, CARL Systems Terry Sullivan, Florida Center for Library Automation Craig Summerhill, Coalition for Networked Information Robert Waldstein, AT&T Bell Labs Kazu Yanaghara, Apple Computer 2. Definition of the testbed project and this group's relationship to other organizations Cliff reviewed how this group came to be and Coalition for Networked Information's (CNI) involvement. Stressed the need to focus efforts on 'compatibility' Jim Michael discussed his views on the relationship to the ZIG. This group 'ZIT' (Z39.50 Implementors Testbed) has an important role in providing a common ground for the discussion of the implementation of the protocol in the current networking environment (TCP). The ZIG's role to define element sets for version 2 (Z39.50 -- 199?) and version 3. The two groups don't need to go in diverse paths but need to work in parallel. Ray sees three bodies: the maintenance agency, ZIG and ZIT. The ZIT should as a result of implementation recommend changes to the ZIG --there must not be a lack of coordination -- don't stop attending the ZIG due to frustration with the ZIG or the maintenance agency. The ZIG is also the de facto body for recommending changes to the maintenance agency. Cliff stressed that he views this group not as a standards writing group but instead, as a group who purpose was to get a working implementation then disband. ZIG to draw up the standard. This group is to share information regarding the implementation of Z39.50 including the ' pain & agony' Interoperability is bigger than than protocol, there are record transfer syntax questions, some broader issues and some narrower There is no clear charter for the ZIG. Some ZIT members want to use the standard not write one. Terry Sullivan raised the question of how this group is perceived within the community. 'We all understand the needs of this group. There is the ZIG has failed view. There is a need for a statement to clarify our purpose here'. ACTION: Cliff will work with Mark Hennenbush to draft a statement . Ray requested that 'terms of reference' be included in the statement, that would link the maintenance agency and clarifies the roles of all groups. Administrative issues: A new mail reflector will be created to handle the administrative matters of this group, discussions of this group will able available to the ZIG (cnitbd@stubbs.ucop.edu came up later that week). General technical discussion still belongs on Z39.50IW@nervm.bitnet. Items to be added to the agenda --'record views' or element set (Peter Ryall) --query type and character sets (Dave Richards) Goals: -- to get all of us running a client and/or server -- get the word out that this really works -- demonstrations. -- Modest demo planned for Net '9 2 between UC and Penn State Profiles are the domain of the ZIG. 3. Review of implementation plans by each testbed member organization Cliff , University of California (Melvyl) The server and client sort of work using ASN.1/TCP. But it doesn't do resource control The server is planned to be up in 1-2 weeks available 24 hours a day. Eric Penn State (Lyons): Almost identical state, server running two months, client not up yet George MCNC/UNC Administration role/ WAIS project . Reported that UNC Chappel Hill have CMS, VMS clients and servers up. Kazu Apple: Nothing running yet, interested in doing the client first Cedric Stanford: Starting with nothing, will work from folio (UNIX system connection) by mid year should have a client, the server will come later. Peter Mead Data Central: Different twist: no clients, special server and most interested in working with NeXt and Apple. Will support the MARC attribute set, need a News attribute set, record syntax questions. Interested in the work of John Kunze on how to request various formats Word Prefect, Microsoft Word, TIFF etc.. The server is targeted for mid June/July Ralph LC: Linked Systems Project (LSP) project Upgrade OSI intend to add on using OSI/CS. Plan on being client and server has the DLA software, need to write to interface 6mos to a year. Focusing on both the client and server. David RLG: Server in test using MARC records intending to provide access to all databases and authority records. Client for testing only, will begin testing over telecommunication loop back, ISO protocols will be supported, working on TCP for Unix client work to start next couple of months. Robert AT&T: Search software will talk to itself. Client up real soon problem though AT&T secure gateway. Time frame client under a month server harder because of security. The type of info that could be available statistics library, all press releases, manual for sale, all in the Unix environment system 5 BSD. Ron UM: Has teamed up with CARL Sean DRA: Library is done that can deal with client/server APDU, working on client should be finished in six weeks, server after that. Bob CARL: Watching till now. Ward Shaw has committed resources, parallel development, hopefully mid-summer with UM Terry Florida: Grant is for OSI, in parallel building interfaces to TCP client /server running under OSI, has DLA code, if IBM opens the session layer that will help, limited service mid July Sara Notis: Customer test, SNA encapsulated/TCP, focus on client mid summer, mainframe can do INITS, server secondary product 6 months beyond clients Bill MIT: Holding breath distributed library systems project. Z39.50 / TCP is the next puzzle piece interested in collaboration ASAP options BRS search and WAIS v.2 Harry TM: Client WAIS is public domain, 10,000 different host contact the server, more people writing clients, want to go to v2, WAIS is successful because there is one implementation. John UCB: client/server v2 CWIS server is up 24 hours a day, on port 210 but, there is only one MARC record. Code will be available. ACTION: Cliff is to develop a matrix of this group's implementation progress. 4. Discussion of what version of the protocol we are implementing Is Z39.50 over TCP true Z39.50? Or is what is the actual architecture of Z39.50 over TCP/IP given the standard was written assuming the service of the lower layers of the OSI stack. Q. Which OSI layer runs on TCP? --No presentation or Session? --OSI/CS; --All 7 Layers? --Layers 5,6,7? Cliff: The simpler the better. His implementation: BER directly over TCP, ISODE in UNIX, and PL/1 routines for coder/decoder Larry LC: Using OSI/CS if possible, but using LSP code now. Dave RLG: Full osi in TCP is much more work, but RLG has been on record as saying Presentation Layer on up, is the right way to do it. Cliff: Yes, Z39.50 is designed for OSI and not having it gives birth to kludges. But the required Presentation Layer bits have not been found in any OSI implementation yet anyway. OSI is the "Millstone From Hell." Therefore we are looking for simplicity in implementation using the current TCP environment, for example BER directly over TCP, ISODE in UNIX, the role of the presentation layer, and the problems encountered by attempting to implement in the IBM OSI/CS world. ACTION: The specification for BER/TCP to become an RFC when written. What protocol? Z39.50 - 199? versio n 2 draft 4 as distributed by NISO for balloting. Ray will supply to anyone who needs a copy. Extension of version 2 which are wanted by this group include --explain --authentication -- external instead of string (some of the issues here are actually external to this discussion, although this work should be tracked by this group, because of its importance to the database vendor community in particular) --authentication challenge/response - External instead of string --Diagnostics -- extending, possibly restructuring --User information field for every APDU, can go in for debugging but this and any other changes that are put in for this project should have the ability to be pulled for production releases if not approved by the ZIG. Much of this experimentation could end up in version 3, as we learn as we go. ACTION: Cliff will maintain a list of the variances use by this group for the purposes of testing. If in the process of implementing other issues of variances arise bring them to the meetings. The discussion then revolved around staying in sych with SR or experimenting. Cliff reiterated that the objective of this group was to have a production base in what looks like 8 months, given the expected time frame for implementation mentioned earlier. Profiling: Will support type 1 and type 101. NO CCL Access control vs. authentication in the init MIT has external instead of string for Kerberos, Mead will be implementing but will challenge on certain databases, Florida has the code but isn't doing anything ACTION: Each implementor is to review the PICS and at the next meeting the PICS will be compiled by someone to identify those specifics which will aid in interoperability. Registration issues: US BIB-1 but be able to recognize the OID for SR BIB- 1. Remember to recognize v.1 Record size? check ZIG min/max, PDU and message size? in general principle plan for full text and images. Two solutions: record segmentation (edge into element set name) or send multiple result set PDU 4 - 32K range ok. 5. Interoperability issues: TCP/IP Discussed elsewhere ASN.1 ASN.1 no major problems except direct reference, indefinite length. Need to accommodate for direct reference of indefinable length. Error messages Error messages still weak- diagnostic set bib-1 will be sorted out in actual implementation. Therefore, start with BIB-1 and add as needed. No convenient way for encoding additional information which will be learned from implementation. Provide a list of diagnostics (server builders) Attribute sets BIB-1 & SRBIB-1 Need an attribute set for explain Mead Data Central needs the 'news' set within the next six-months. John reviewed the Dynamic attribute set for CWIS or other non- bibliographic databases, and the record format. He will have a new draft of the dynamic attribute set INFO-1. Peter said he could use that type of generic view 'brief' vs. 'full' What are the core attributes? v3 attach attribute set to each term? (the multiple database question) It was purposed that the server should put up list of what it supports as part of an explain function. For the purpose of demo need to list of attributes. John suggest it is just good practice to code defensively. We can use the list as crutches in the short term, not just attributes but in what combination Record transfer syntaxes 1) MARC (or MARC like) + sometimes holdings 2) Abstracting and Indexing (A&I) 3) Full text 4) Image files 5) WAIS headline format 6) ASN.1 brief (card catalog like) WAIS headlines and brief bib to be merged. David RLG full MARC - not sure on the holdings, and the character set is anything that's in the record. LC Authority records are a separate file. A MARC bib record and include holding and circulation records LC is coding USMARC no holdings Sara NOTIS: MARC with holdings, really need a brief view and some indication as to what the hit was against. Accommodated in brief bi b. Need to do/have to have abstract syntax. 1 , MARC (Can be used for just holdings ) MARC + Holdings, Short BIB (New record syntax see Mark H notes from ZIG) could also be used for the WAIS brief headline. WAIS could also use tagged list. General (composite data types) could go into info-1 Real near term records, MARC, Brief, Some image (FITS, TIFF, FAX) emphasis in TEXT IMAGE questions Monochromatic draft RFC for TIFF, the formatting issues for image transfer should be pushed back to the server. IETF rev 822- multimedia, pick type of image format, preferred record syntax, record syntax, -- now in element set name. Record syntax is generic images came as a field. Near term unreasonable to move down an image transfer format. Format type (big bucket of bits) INFO-1 as attribute set. Fairly easy to deal with images nail down detail later. TEXT shipping. text, text &, text-lines, Word perfect, TROFF, TEX? Short term solutions not in this level of detail yet. What equals a CR? Build on standard or quasi standards none defined and implemented. Puts allot of load on the client -- will let us learn . How reasonable is it to put that load on the client? A& I Notis maps to MARC. Review USMARC, OID Brief record (summary) WAIS & document description ASN.1, title, author, size, date. Primitive tagged format (TIFF, WP) octet, summary record ASN.1. TEXT ASCII with and with out line feed. Bundled (text, image, sound) variant of John's INFO-1 George Brett raised a question about what happens with something like the Duke project with WAIS and Quicktime for the art database? Specialized problems (e.g. WAIS model) 6. Interopability testing plans Who can test against who, an when Character set issues (in terms) assume ASCII now. The element set issue brought up earlier by Mead Data Central can be resolved in the dynamic attribute set and brief record. Test cases/ plan batch mode APDU available at DLA. Test suites? Larry at LC has test scripts if any one wants them. How to find servers and where they are? Post domain names, IP and port. -- John infocal.berkeley.edu port 210. (note this is a change from what was discussed at the meeting) RLG to have testable services in March, tricky question of access for both RLG and Mead they will have to set up specific databases for testing. Base set: search, retrieval, MARC or headline record won't open to the public, for purposes of testing and development to this group only. Issues involved in debugging and access for testing purposes Debugging tools and facilities. DLA has a way to dump out APDU, cross organization debug. Need someway to send back -- 'shut down the connection' error PDU. Terminate in v3. Extension - error PDU - shut down system -- applications and abort = TCP abort. The error PDU visible string/Octet string. Resource control and the state tables Looks like by April there will a reasonable number of servers with a variety of clients 7 Plans for demos and the conect-a-thon Specific demo's planned Net '92 What other communities? EDUCOM Oct., CNI task Force Nov., CAUSE Dec., ALA mid-winter Jan., LITA Sept 13 Denver CARL to look into the internet drop) 8. Schedule for future meeting The next meeting has been scheduled for March 12 in the area of the San Francisco International Airport. The exact location will be announced later. The next meeting will be to firm up the details of implementation. Other meetings will be scheduled as needed. Note: I would like to thank Bill Cattey, for sharing his notes with me. Cecilia