Views on Issues Relevant to Data Sharing
on Computer Networks
Introduction
The formation of a committee to address the problems of achieving
data sharing on the ARPA Network, as suggested by Arie Shoshani
(RFC #140) is desirable at this point of network development. We
concur with Shoshani's ideas (presented in an introductory paper
to the network data sharing meeting, scheduled for Tuesday, May 18)
and believe that purpose of the committee should be -
a) to classify the issues involved and to propose various
approaches;
b) to integrate the hitherto independent network activities
that address problems in the area of data sharing, and;
c) to set up and coordinate appropriate experiments to test
the services developed and to evaluate alternative
approaches.
This position paper is intended to augment Shoshani's as a basis
for discussion at the data sharing meeting. No attempt is made
to discuss specific means of implementation since many approaches
to data handling problems are possible and have been proposed.
Rather, our viewpoint on what the committee's role should be in
giving some cohesion to various existing implementations is
presented.
[Page 1]
Our Views
One approach to achieving data sharing on the ARPA Network can
be thought of as having three stages, which roughly correspond to
the modes of use or operation. Within each stage are various levels
of development required to get to the next stage. This development
is not necessarily sequential. A description of the three stages
follows.
Stage 1: Data handling services are provided at various Hosts.
The user talks directly to the serving Host (via TELNET
or by addressing a known socket) to explicitly access
the service. This mode of operation corresponds to
Bhushan's category of "direct" usage (RFC #114). The
data services provided by the serving Host range from
simple ones, such as White's file transfer system (RFC #122)
to sophisticated systems such as the CCA's data machine
(NIC 5791 and 6706).
Stage 2: The user has access to an intermediate process or data
control facility* that routes his requests for a particular
data service to the serving system. The user must explicitly
identify the data services to the used. This mode of
operation corresponds to Bhushan's category of "indirect"
access. The data control facility provides the necessary
control commands, data transformations, and accessing
methods. A single request would include the use of several
interacting services. For example, Heafner's Data
Reconfiguration Service (RFC #l38) could be used in
conjunction with the use of CCA's data machine.
_______________
*The data control facility is not necessarily located at his local
Host. Such a facility may exist on from one to all Host (i.e.,
ranging from centralized to completely distributed).
[Page 2]
Stage 3: The user treats the network as a single resource and is
unconcerned with the location of the services, data files,
etc. All references are by name. In this mode of opera-
tion, the data control facility can function as a referral
center for data service requests by using the most ap-
propriate data service available and by automatically
combining the use of several services that may be needed
to satisfy a request. For example, data could be retrieved
from several files, each managed by a different data
management system. The data control facility must be
cognizant of the location of data files, their structure,
data management system capabilities, etc.
Some approaches to the design of the data control facility have
been suggested by Shoshani, notably the integrated data management
system (IDMS) and the unified data management system (UDMS). The
notion of the network machine (RFC #51) is closest to the capabilities
one would see in Stage 3.
Relevant Areas of Development
The data control facility can range anywhere from a simple inter-
face to an intelligent front-end processor to a network-wide re-
ferral system. In any case, a common means is desirable for
handling applications such as file transfer, on-line update and
retrieval of data, information gathering and reporting, and program
access to data. To attain this end, a few of the areas in which
developments will be required include:
a) a data description language, permitting the user to define
the physical structure of files, to define logical files,
and to categorize data fields for name referencing. The
language should be designed to facilitate the resolution of
physical discrepancies in data and file structures. The
user should be able to superimpose logical restructuring of
data without any change in the physical structure.
[Page 3]
b) a control or access language that can be mapped into
various data management languages. Considered here is
Shoshani's suggested two-level approach with perhaps a
meta-language implementation to facilitate conversions
among already existing languages.
c) methods for managing and merging distributed data, search
mechanisms for file directories, error recovery techniques,
etc.
Independent ARPA Network activities that in effect constitute
Stage 1 have touched on these areas and should be incorporated into
the overall data sharing scheme such that all of the isolated
pieces are compatible. For example,
a) the data reconfiguration service (RFC #138) would be
invoked by the data control facility whenever data transformations
are required.
b) the file transfer protocol (RFC #114, #122)
should be consistent with other data handling services.
c) CCA's data machine should be a subset or part of any data
control facility. The network data language and set of data
management services that they plan to implement can perhaps be
adopted network-wide.
d) the network machine concept (RFC #51) for defining the pro-
gram and data environments should be resurrected. The data control
facility should be a subset of a network machine architecture.
Some other relevant topics include NIL (RFC #51), DEL (RFC 5), the
notion Of MYLOCAL n, YOUR LOCAL n, and STANDARD n (RFC #42), user
level protocol objectives as described in RFC #76 and #91.
[Page 4]
Experimentation and Testing
---------------------------
As data services are developed on the network, a coordinated
effort is desirable
a) to exercise individual implementations to see
if they work, both alone and in conjunction with
other data services, and
b) to evaluate alternative approaches.
Some examples of experimentation to test data services follow:
1. File Transfer Protocol
The file transfer protocol should be used to
manipulate data files controlled by various
systems.
2. Data Transfer to Data Computer
The ability to transfer existing data bases and
their structures onto the data computer should be
demonstrated.
3. Data Restructuring
The ability to define logical restructuring of
data for users needs which would be accessible by
name should be demonstrated. The original physical
structure would be maintained.
4. Data Transformation
The ability to access various data management
systems on the network without the user being
concerned with the data transformation involved
should be demonstrated. Necessary calls to forms
available on the Data Reconfiguration Service
should be handled automatically and should be
transparent to the user.
[Page 5]
5. Data Consistency
Problems of maintaining consistency when duplicate
copies of a data file exist and updates to the file
are made should be investigated. Automatic use of
file transfer protocol and DRS to generate new
duplicate copies should be included.
6. Data Privacy
Access controls for privacy Of data files in the
network environment should be designed and evaluated.
This includes controls on parts of distributed files.
Our recommendation is that the committee on data sharing be
responsible for coordinating development in these areas, for
attempting to maintain consistency among data services, and for
testing services in a series of experiments as they are implemented.
[ This RFC was put into machine readable form for entry ]
[ into the online RFC archives by BBN Corp. under the ]
[ direction of Alex McKenzie. 12/96 ]
