Network Working Group D. B. McKay
Request for Comments: 304 IBM
NIC: 9077 February 17, 1972
Categories: D3, D4, D7
Obsoletes: none
Updates: none
A Data Management System Proposal
for the ARPA Network
Introduction
This proposal is being written to facilitate discussions on a design
for a network Data Management System. It is not intended to be a
complete and exhaustive design for the ultimate protocol allowing
users to share data easily, but a frame work that will allow us to
recognize and develop the necessary tools in a unified manner
enabling the network to manage its resources to the best advantage to
the user.
The fundamental intent here is not to try and solve an impossible
problem, but to bring a necessary service capability to the user that
will enable him to carry out applications that hitherto he has not
been able to do. The intent is to be consistent with every other
major function that has been developed in the network, i.e., NCP -
2nd level protocol, Telnet, and the Form Machine. The Data
Management Service or Data Control Facility (DCF) will do the same
thing only on a high level of application building on those tools
that have already been developed in the network.
Data that is referred to and transmitted in this System will be
considered a special class of data that is called network data. That
is, it is named and characterized through a network datalanguage and
all pertinent information as to where it can be located and what its
structure is kept in a network catalog. Access to the data for its
actual transmission will be done through NCP socket addressable
routines in a manner similar to the way in which the SMFS at Santa
Barbara works. It is feasible that the SMFS will become an active
resource utilized by the DCF.
System Overview
There are six functionally and logically distinct areas that are
identifiable in the Network Data Service (Figure 1), with
subfunctions that can be categorized and discussed.
McKay [Page 1]
RFC 304 A Data Management System Proposal for ARPA February 1972
1. The user interface to the DCF. In an interactive environment
such as the ARPA network, this interface would be serviced by
Telnet supporting the local user at his terminal directing the
request to the DCF. The DCF in this case would be a
specialized server task.
2. The DCF or that functional unit responsible for coordinating
all the activity of the Network Data Service. It also houses
the interfaces to all other functions.
3. The Network Catalog or Directory which contains all information
about network data.
4. The Data Reconfiguration Service or Form Machine that would be
called on when data translation or reconfiguration is needed.
This would be invoked automatically, when possible, by the DCF
and would remove this responsibility from the user. For more
specialize translation, however, the user will still be able to
write programs for, and execute them on, the Form Machine.
5. The remote DCF or DCF' would contain enough function to
recognize the request being made of it by the DCF It would be a
server task to the DCF.
6. File xfer protocol would be a function that the DCF and the
DCF' would initiate as the means to control data transfer in
the network.
A more detailed discussion of each of these areas appears in the
following sections.
User Interface
It was stated in RFC146 that the DCF should handle all network
resources as a single resource and utilize it as best it can. This
statement was also meant to incorporate the Data Computer and Unicon
storage as part of this resource. The extent to which this can be
done is an open question but the use of the Data Language developed
by CCA would provide a consistent interface to the user utilizing
these network services and possibly facilitate the use of the Data
Compiler by the DCF.
It should be pointed out at this time that the DCF is a logical
function that can reside anywhere including on the Data Computer.
The user should be allowed to enter all command and updates
interactively to the DCF. The DCF will be a serving user process
that will interface to the Telnet Server routine. The actual data of
the terminal transmissions will be the commands and data the user
will be transmitting to the DCF. By adopting the Telnet protocol as
the initial user interface, the DCF can be accessed by all the users
with Telnet.
McKay [Page 2]
RFC 304 A Data Management System Proposal for ARPA February 1972
The actual user commands and data itself is an area that requires
more investigation. The following comments offer suggestions as to
what a final data language and manipulation language should do.
There are at least two logically distinct functions that must be
performed. The actual defining and redefining of the data, and the
request for service such as catalog entry and request for
information. This proposal is not intended to provide access to
every data base in the network; instead it is aimed at those files
that are catalogued and known to the DCF in the manner analogous to
the Data Computer's knowledge of its Data Base.
The following Data Description Concepts described in detail in CCA's
Data-language are also useful in the DCF. First of all, the Data
Containers are groups of nested boxes. The box represents the data
or other data containers that are kept in the box. It represents a
named set of locations in some storage medium. There are also
several types of data container such as STRING, INT, REAL, PTR,
ARRAY, LIST, STRUCT, and MIX. Finally, each of these containers can
be named. The name can also qualify the item of interest by a
concatenation of names to reflect the logical nesting of the
containers.
Although storage and retrieval mechanisms should be the same as those
proposed in the DL, initially it should not be necessary to implement
all the functions that filter and manipulate the data. For example,
in an initial implementation of a DCF it would not be necessary to
provide the user with relational, boolean and computational
operators. Users specifically interested in this type of service
could be directed to use the Data Computer under his own initiative,
or as a service of the DCF.
Several of the operators specified in the language are important and
must be considered in the DCF.
The following list represents key operators for a DCF with a brief
description of what each function is. For a more detailed discussion
of each statement the reader should read the Datalanguage report by
CCA.
The assign operation places value into the containers with the
containers being single items or referencing other containers.
Subscripting allows selection by element number. It is a powerful
tool for specifying, in large containers of data, the reference and
transmissions of only the necessary parts. Files can be subsetted by
McKay [Page 3]
RFC 304 A Data Management System Proposal for ARPA February 1972
containerization referencing fields and records which can be further
classified by subscripting. This subscripting function can be
further extended to include the DL's virtual list concept.
Maintenance of the files must be provided with the delete and add
function applied to the container referenced data.
A second consideration in a catalog is the question of how feasible
it would be to keep back-up or duplicate copies of network files.
This of course raises the question of a multiple copy update
protocol. I feel the discussion and development of this protocol,
although important, can be postponed in lieu of keeping multiple
copies of files that are primarily read-only files.
For experimental reasons the DCF should have at least one data base
that is kept at different locations - possibly NIC, with the
capability of access anyone of them in the event of system failure at
other locations. This is an important point, it exploits one of the
major advantages of a computer network namely more reliable data
accessibility.
Finally, the actual location of the network directory is an
interesting question. In the interest of reliability it should be
kept at multiple locations. The network directory can be logically
separated into two segments. The local directory and distributed
directory. Both parts refer to network data. The local segment is
kept up-to-date relative to the network data that resides on that
system. The network segment records the location of files that are
duplicated on other file systems and system pointers to references
made of remote single system files.
Updates can be made to the network segment on a periodic basis.
These updates will reflect changes in the local segments. If we
consider "read-only" files distributed initially and local segments
reflecting the changes in local files, the need for simultaneous
update of multiple copies and network segments of the catalog becomes
much less critical. Based on the two segment approach to the network
directory it seems most convenient to keep copies on all systems that
have localized network data. This would include a catalog on the
Data Computer.
Data Conversion (Form Machine)
The Form machine represents an essential network function that can be
invoked by the DCF when necessary. The Form Machine would be used in
the same manner it was intended to be used only now the DCF would
intervene in place of the user. This would represent a common
McKay [Page 4]
RFC 304 A Data Management System Proposal for ARPA February 1972
interface to the network for the user. Having the user use the
Datalanguage for file transfer and conversions would mean the DCF's
management of his Form machine Services.
The motivation behind the Form machine is consistent with a service
that should be provided in a network data sharing facility, namely,
application programs require different formats from program to
program and the network should adapt to the individual program
requirements. This is also true of console configurations and
machine dependent data.
The modus operandi of the service is descriptions of data are
supplied by the application programmer in forms that the service
stores by name. In the case discussed here the DCF would invoke the
data transformation on the network data stream by calling the forms
by name. These would be a standard set of forms for machine
dependent data that would be written as part of a general
implementation of the DCF and would be invoked when necessary by the
DCF.
There are three conceptual connections to the Data Reconfiguration
Service (DRS).
1. The critical connection between the originating user an the
DRS. In this case it would be the DCF. This raises the
question of how the user would communicate with the Form
Machine. He could use normal procedures and the directory
to the DRS by Telnet or he could allow the DCF establish a
connection for him with his defined forms cataloged in the
network catalog automatically.
2.-3. The other two connections are between the user process and
the serving process which are made by the DRS through the
NCP.
Since the Form Machine represents an invaluable service to the
network it is imperative that it reside in several locations with
user named forms available at each DRS location. This will ensure
available service when needed. However, having the DRS invoked by
the DCF raises two interesting areas of investigation.
The first is the question of the common interface that the network
presents to the user. If the Datalanguage is to be the common
interface, then is it practical and feasible for a mapping service to
be performed by the DCF that will convert Datalanguage statement into
the proper parameters and statements necessary to the form machine?
This is an area that has to be discussed and further investigated. I
would encourage anyone to submit an RFC on this subject.
McKay [Page 5]
RFC 304 A Data Management System Proposal for ARPA February 1972
The second question is a simple one and has to do with the question
of the file transfer protocol and whether or not it is sufficient to
handle the requirements for the form machine connections from a
remote location.
File Transfer Protocol (FTP)
The objectives of the FTP are to promote sharing of files, encourage
implicit (without explicit login) use of computers, and to shield the
user from variations in file and storage systems of various hosts.
The use of these related operation in the FTP by the DCF allows us to
extend these ideas to the restructuring and subfile level that was
discussed in the user interface section.
In addition, request can be made to any DCF on any system to retrieve
a file and transfer it to any other system. This means not only that
implicit use of systems be encouraged but the user will be removed
from the burden of explicitly having his system linked to the system
transferring the file. For example, if someone is running an
analysis program at BBN which may also have a DCF. A request for a
file to be shipped to BBN from the Santa Barbara 75 could be made to
the DCF at BBN without the user having to communicate with the 75
directly.
The sending and receive connection would be initiated by the DCF with
the logical link between the two systems obeying the FTP. The only
modification I can see in the FTP that would be necessary is an
acknowledgement to the commands sent to the sending and receiving
sites by the DCF. In addition, an acknowledgement to the end of file
indication would be sent to the sending system and to the DCF. The
rename from, rename to, delete and list request would be transmitted
by the DCF directly with all acknowledgments being returned to it.
The remote DCF and DCF' mentioned earlier would recognize and handle
all the FTP messages. In addition, it would recognize requests being
made for a particular container or subset of the data. It should be
able to recognize the information given to it, access the data
requested and be able to strip off the necessary information
requested and transmit it.
The complexity of the DCF' would depend on the amount of functional
capability that was incorporated into the network portion of the
Data-language.
McKay [Page 6]
RFC 304 A Data Management System Proposal for ARPA February 1972
Conclusion
This paper is intended to promote ideas and discussion in all of the
areas mentioned. The principle outcome is to start a coordinated
specification and implementation effort to provide data sharing in
the network.
,---------------.
| |
| DCF' |
Sending | |
Host `---------------'
\ ^ /
\_____________|__________/
|
/ | ^
File / | |
Xfer / |
Protocol / DRS | |
/ Protocol |
/ | |
/ |
,----------. /\ ,---------./ /\ ,------------|--.
| | / \ | | / \ | |
| TELNET |---\ / ---| DCF |---\ / ---| DRS | |
| | \/ | | \/ | |
`----------' | |\ `------------|--'
^ `---------' \File |
| ^ | \Xfer DRS | |File
| | | \Protocol Protocol | Xfer
v | v \ | |Proto-
,----------. ,---------. \ | col
| USER | | | \ | |
| TERMINAL | | CATALOG | \ | v
| | | | \ _______________________
`----------' `---------' / | \
/ v \
Receiving ,---------------.
Host | |
| DCF' |
| |
`---------------'
Network Data Service
System Overview
Figure 1
McKay [Page 7]
RFC 304 A Data Management System Proposal for ARPA February 1972
The keyword statements of the language are important for data
manipulation and transfer. These keywords will initiate entry of
information into the net catalog and access the physical data located
at the various systems. Most of these keyword commands would be
directed to the remote systems as part of the file transfer protocol.
Some examples of the keywords incorporated by the DL are, CREATE,
DELETE, OPEN, TRANS, CLOSE, and DEFINE.
Network Catalog and Directory
The actual structure of the network catalog should be fairly
straight-forward. It will contain all the information necessary to
retrieve data files and designated subsets of those files. Initially
the catalog need not contain all the information one would hope to
have such as authorization for use, access, or update, therefore it
is imperative that the catalog be an open ended structure that can be
easily added to.
The primary purpose of the network catalog will be to store all
network data file structure information that the user has entered via
the Data-language. It must also contain an indication of how the
users logical description of his file is associated with the actual
physical file and location. This physical information must contain
the proper pointers and addresses to actually retrieve the data.
Since the class of files we will be dealing with are network files
that will be accessible by the network user function such as SMFS,
the addressing information can be pathnames as suggested in the
Rename Convention in the file transfer protocol.
[ This RFC was put into machine readable form for entry ]
[ into the online RFC archives by Mirsad Todorovac 11/98]
McKay [Page 8]