Network Working Group C. Lynch
Request for Comments: 1729 University of California
Category: Informational Office of the President
December 1994
Using the Z39.50 Information Retrieval Protocol
in the Internet Environment
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.
Summary
This memo describes an approach to the implementation of the
ANSI/NISO Z39.50-1992 Standard for Information Retrieval in the
TCP/IP environment which is currently in wide use by the Z39.50
implementor community.
IntrodUCtion
Z39.50 is a US national standard defining a protocol for computer-
to-computer information retrieval that was first adopted in 1988 [1]
and extensively revised in 1992 [2]. It was developed by the National
Information Standards Organization (NISO), an ANSI-accredited
standards development body that serves the publishing, library, and
information services communities. The closely related international
standard, ISO 10162 (service definition) [3] and 10163 (protocol)
[4], colloquially known as Search and Retrieve or SR, reached full
International Standard (IS) status in 1991. Work is ongoing within
ISO Technical Committee 46 Working Group 4 Subgroup 4 to progress
various extensions to SR through the international standards process.
The international standard is essentially a compatible subset of the
current US Z39.50-1992 standard. Z39.50 is an applications layer
protocol within the OSI reference model, which assumes the presence
of lower-level OSI services (in particular, the presentation layer
[5]) and of the OSI Association Control Service Element (ACSE) [6]
within the application layer.
Many institutions implementing this protocol chose, for various
reasons, to layer the protocol directly over TCP/IP rather than to
implement it in an OSI environment or to use the existing techniques
that provide full OSI services at and above the OSI Transport layer
on top of TCP connections (as defined in RFC1006 [7] and
implemented, for example, in the ISO Development Environment
software). These reasons included concerns about the size and
complexity of OSI implementations, the lack of availability of mature
OSI software for the full range of computing environments in use at
these institutions, and the perception of relative instability of the
architectural structures within the OSI applications layer (as
opposed to specific application layer protocols such as Z39.50
itself). Most importantly, some of these institutions were concerned
that the complexity introduced by the OSI upper layers would outweigh
the relatively meager return in functionality that they were likely
to gain. Thus, for better or worse, the decision was taken to
implement the Z39.50 protocol directly on top of TCP (with the
understanding that this decision might be revisited at some point in
the future).
During 1991-1993, a group of implementing institutions agreed to
participate in the Z39.50 Interoperability Testbed project (sometimes
referred to by the acronym "ZIT") under the auspices of the Coalition
for Networked Information (CNI). Their primary objective was to
encourage the development of many interoperable Z39.50
implementations running over TCP/IP on the Internet. By mid-1993, a
number of independent Z39.50 implementations were operational and
able to interoperate across the Internet.
The Library of Congress, in its role as the Z39.50 Maintenance Agency
for NISO, maintains a registry of the implementors [8], which
includes members of the Z39.50 interoperability testbed.
This document describes implementation decisions by current
implementors of Z39.50 in the Internet environment. These have been
proven within the ZIT project and are being used by most of the
members of the Z39.50 Implementors' Group (ZIG), an informal group
that meets quarterly to discuss implementation and interoperability
issues and to develop extensions to the Z39.50 protocol targeted for
inclusion in future versions of the standard. Intended as a guide for
other implementors who seek to develop interoperable Z39.50
implementations running over TCP/IP, this document focuses on issues
related to TCP/IP, and it does not address other potential
interoperability problems or agreements that have been reached among
the implementors to address these problems. It does include a few
notes about extensions to the existing Version 2 protocol that are
being used in the implementor community which have interoperability
implications. Potential implementors of Z39.50 should subscribe to
the Z3950IW LISTSERV [9] to oBTain information specific to the Z39.50
protocol and extensions under development as well as details of
current implementations.
Except where otherwise noted, the version of Z39.50 discussed here is
ANSI/NISO Z39.50-1992, sometimes called Z39.50 Version 2 (the
obsolete original version is referred to as Z39.50-1988 or Z39.50
Version 1). The approach defined should also be applicable, perhaps
with some minor changes, to future versions of the Z39.50 protocol,
and specifically to Version 3 which is currently under development.
This document will probably be updated to address new versions of the
base Z39.50 protocol as they become stable.
Encoding
The Z39.50 standard specifies its application protocol data units
(APDUs) in Abstract Syntax Notation One (ASN.1) [10]. These APDUs
include EXTERNAL references to other ASN.1 and non-ASN.1 objects such
as those defining record transfer syntaxes to be used in a given
application association.
The standard Basic Encoding Rules (BER) [11] are applied to the ASN.1
structures defined by the Z39.50 protocol to produce a byte stream
that can be transmitted across a TCP/IP connection. The only
restriction on the use of BER to produce this byte stream is that
direct, rather than indirect, references must be used for EXTERNAL
objects. This is necessary because there is no presentation context
in the TCP/IP environment to support indirect reference. A Z39.50
implementation developed according to this specification and running
over TCP/IP should produce a valid byte stream according to the
Z39.50 standard, in the sense that the same byte stream could be
passed to an OSI implementation. However, not all byte streams that
can be produced by applying BER to the APDUs specified in the Z39.50
standard in an OSI environment will be legitimate under this
specification for the TCP/IP environment; this specification defines
a subset of the possible byte streams valid in a pure OSI environment
which excludes those using indirect reference for EXTERNAL objects.
All other BER features should be tolerated by Z39.50 implementations
running over TCP/IP, including the ability to accept indefinite
length encodings, although it is preferable that implementations do
not generate such encodings since they have caused problems for some
ASN.1/BER parsers. It should also be noted that at least to the best
of the author's knowledge, there are no implementations at present
that use ASN.1/BER representations of floating point numbers;
instead, integers with scaling factors have been used for these
purposes. It should also be noted that Z39.50 version 2 does not
really address character set encoding issues; these questions, and
their interactions with ASN.1/BER support for multiple character
sets, are under active discussion as part of the effort to develop
Z39.50 version 3.
Connection
In the Internet environment, TCP Port 210 has been assigned to Z39.50
by the Internet Assigned Numbers Authority [12]. To initiate a Z39.50
connection to a server in the TCP/IP environment, a client simply
opens a TCP connection to port 210 on the server and then, as soon as
the TCP connection is established, transmits a Z39.50 INIT APDU using
the BER encoding of that INIT APDU as described above.
Implementors should be aware that there is a substantial installed
base of implementations of the Wide Area Information Server (WAIS)
system. The original versions of this software employed Z39.50
Version 1 with some extensions. Z39.50 Version 1 did not use BER
encoding and Z39.50 Version 1 INIT APDUs look very different from the
INIT APDUs of Z39.50 Version 2. Implementations of Z39.50 should at
least be prepared to reject gracefully WAIS-type INIT APDUs. Some
implementations recognize such INIT APDUs and revert to the Z39.50
Version 1 variant used in WAIS upon encountering them, thus providing
backwards compatibility with the existing base of WAIS clients and;
the usual means of checking for a WAIS, as opposed to Z39.50 Version
2, client is to see if the first byte sent on the connection is an
ASCII zero, which indicates a WAIS client. (In version 1 of WAIS,
bytes 0-9 of the first PDU contain an ASCII packet length; the lower
case ASCII string "wais" appears starting at byte 12.) Work is
currently underway to specify a WAIS profile for use with Z39.50
version 2 [13]; it is eXPected that this will be issued as a Z39.50
Applications Profile through the NIST OIW Library Automation Special
Interest Group. This profile is expected to be compatible with the
layering defined in this RFC.
Service Mappings
The Z39.50 standard maps Z39.50 services onto a variety of
association control and presentation layer services. Connection
establishment has already been discussed. The other two association
control services that are relevant to Z39.50 are ABORT and RELEASE.
The mapping of the RELEASE service to a standard TCP CLOSE is
straightforward. The Z39.50 protocol itself does not, in the current
version, include a Z39.50 CLOSE APDU. When the client has completed
its interaction with the server, it calls the IR-RELEASE service,
which is directly mapped to association control's orderly association
release. In the TCP/IP environment, the client should simply initiate
a TCP CLOSE. The mapping for association abort is more complex,
partially because some TCP/IP implementations cannot distinguish a
TCP reset from the other side of the connection from other events. To
accomplish an abort (that is, a mapping of the IR-ABORT service in
the Z39.50 protocol) in the TCP/IP environment, client or server need
only terminate the TCP connection either via TCP ABORT or TCP CLOSE.
Real-world implementations need to be prepared to deal with both TCP
ABORT and CLOSE anyway, so this approach presents no additional
problems, other than the somewhat ambiguous nature of the type of
association termination.
It is expected that Z39.50 Version 3 will include a termination
service which will involve an exchange of Z39.50 CLOSE APDUs,
followed by an association RELEASE (which would presumably, in the
Internet environment, be mapped to a TCP CLOSE). This new termination
service is expected to support both graceful and abrupt termination.
Of course, robust implementations will still need to be prepared to
encounter TCP CLOSE or ABORT.
Service mappings for the transmission of data by client and server
(to the presentation layer P-DATA service) are trivial: They are
simply mapped to TCP transmit and receive operations. TCP facilities
such as expedited data are not used by Z39.50 in a TCP environment.
Contexts
At the point when the TCP connection is established on TCP port 210,
client and server should both assume that the application context
given in Appendices A and B of the Z39.50-1992 standard are in place.
These are the ASN.1 definitions of the Z39.50 APDUs and the transfer
syntax defined by applying the BER to these APDUs.
Implementations can reasonably expect that the diagnostic set BIB-1
is supported, and, if resource control is being used, the resource
report format BIB-1 is supported as well.
In the absence of a presentation negotiation mechanism, clients and
servers should be cautious about using alternative attribute sets,
diagnostic record formats, resource report formats, or other objects
defined by optional EXTERNALs within the Z39.50 ASN.1, such as
authentication parameters, unless there is known to be prior
agreement to support them. Of course, either participant in an
association can reference such an object by object ID in an APDU, but
there is no guarantee that the other partner in the association will
be able to understand it. Robust implementations should be prepared
to encounter unknown or unsupported object IDs and generate
appropriate diagnostics. Over time, the default, commonly known pool
of object IDs may be expanded (for example, to support authentication
parameters).
Implementors should refer to the document [14] issued by the Z39.50
maintenance agency in June 1992 for more details on the assumed
contexts and object identifiers.
Record syntaxes present a serious practical problem. In the OSI
environment, the partners in a Z39.50 association are assumed to
agree, either through presentation negotiation as part of association
establishment, or later, dynamically, as part of the PRESENT process
(through the use of the alter presentation context function at the
presentation layer), on which record syntaxes the two entities
commonly know. There is a preferred record syntax parameter that can
be supplied by the client to guide this negotiation. A number of
registered record syntaxes exist; some are based on ASN.1 and others
use formats such as the MARC standard for the interchange of machine
readable cataloging records which predate ASN.1, but are widely
implemented. In the TCP/IP environment, if the server cannot supply
the record in the preferred syntax, it has no guarantee that the
client will understand any other syntax in which it might transmit
the record back to the client, and has no means of negotiating such
syntaxes.
Several proposals have been suggested to solve this problem. One,
which will likely be part of Z39.50 Version 3, is to replace the
preferred record syntax parameter with a list of prioritized
preferred syntaxes supplied by the client, plus a flag indicating
whether the server is allowed to substitute a record syntax not on
the list provided by the client. The currently proposed ASN.1 for
this extension is upwards compatible with Z39.50 Version 2, although
the details are still under discussion within the Z39.50
Implementor's Group. As the Version 3 ASN.1 becomes stable in this
area, Z39.50 servers are encouraged to accept the extended ASN.1 for
generalized preferred record syntax. The extensibility rules for
Z39.50 negotiation let clients and servers negotiate the use of
Z39.50 Version 2 plus the generalized preferred syntax feature from
Version 3. Thus, a client could support the generalized preferred
record syntax, propose its use to any server, and, if the server
rejects the proposal, revert to the Version 2 preferred syntax
feature.
A second alternative (not incompatible with the Version 3 extension)
would be to adopt a convention for TCP/IP implementations that the
server not return a record in a syntax not on the preferred record
syntax list provided by the client. Instead, it would return a
diagnostic record indicating that a suitable record transfer syntax
was not available. This strategy could be viewed as simply
implementing a subset of the Version 3 solution, and should be
considered by implementors of servers as a possible interim measure.
Other Interoperability Issues
Version 3 will include an "other" data field in each APDU, which can
be used to carry implementation-specific extensions to the protocol.
A number of implementations are already employing this field, and
interoperable implementations might be wise to include code which at
least ignores the presence of such fields rather than considering
their presence an error (in contravention of the standard).
References
[1] National Information Standards Organization (NISO). American
National Standard Z39.50, Information Retrieval Service
Definition and Protocol Specifications for Library Applications
(New Brunswick, NJ: Transaction Publishers; 1988).
[2] ANSI/NISO Z39.50-1992 (version 2) Information Retrieval Service
and Protocol: American National Standard, Information Retrieval
Application Service Definition and Protocol Specification for
Open Systems Interconnection, 1992.
[3] ISO 10162 International Organization for Standardization (ISO).
Documentation -- Search and Retrieve Service Definition, 1992.
[4] ISO 10163 International Organization for Standardization (ISO).
Documentation -- Search and Retrieve Protocol Definition. 1992.
[5] ISO 8822 - Information Processing Systems - Open Systems
Interconnection - Connection Oriented Presentation Service
Definition, 1988.
[6] ISO 8649 - Information Processing Systems - Open Systems
Interconnection - Service Definition for the Association Control
Service Element, 1987. See also ISO 8650 - Information Processing
Systems - Open Systems Interconnection - Protocol Specification
for the Association Control Service Element, 1987.
[7] Rose, M., and D. Cass, "ISO Transport Layer Services on Top of
the TCP, Version 3", STD 35, RFC1006, Northrop Research and
Technology Center, May 1987.
[8] Registry of Z39.50 Implementors, available from the Z39.50
Maintenance Agency (Ray Denenberg, ray@rden.loc.gov)
[9] To subscribe to the Z39.50 Implementor's Workshop list send the
message: SUB Z3950IW yourname to: LISTSERV@NERVM.NERDC.UFL.EDU
(or NERVM.BITNET). Current drafts of the Version 3 Protocol
document are available through the Library of Congress GOPHER
server, MARVEL.LOC.GOV.
[10] ISO 8824 - Information Processing Systems - Open Systems
Interconnection - Specifications for Abstract Syntax Notation One
(ASN.1), 1987
[11] ISO 8825 Information Processing Systems - Open Systems
Interconnection - Specification of Basic Encoding Rules for
Abstract Syntax Notation One (ASN.1) 1987
[12] Reynolds, J., and J. Postel, "Assigned Numbers", STD 2, RFC1700,
USC/Information Sciences Institute, October 1994.
[13] WAIS Profile of Z39.50 Version 2, Revision 1.4, April 26, 1994,
available from WAIS Inc.
[14] Registration of Z39.50 OSI Object Identifiers (Z39.50-MA-024),
available from the Z39.50 Maintenance Agency (Ray Denenberg,
ray@rden.loc.gov).
Security Considerations
This document does not discuss security considerations. However, it
should be noted that the Z39.50 protocol includes mechanisms for
authentication and security that implementors should review.
Author's Address
Clifford A. Lynch
University of California, Office of the President
300 Lakeside Drive, 8th Floor
Oakland, CA 94612-3550
Phone: (510) 987-0522
EMail: clifford.lynch@ucop.edu
