Network Working Group A. Shoshani
Request for Comments: 144 SDC
NIC: 6729 30 April 1971
Data Sharing on Computer Networks
The enclosed is an introdUCtory paper for the meeting which will be
held in Atlantic City as part of the ARPA Network meetings. The
schedule for the meeting will be published soon by Steve Crocker.
The Agenda of the meeting will include:
a. Presentation of the introductory paper.
b. Open discussion to exchange comments and ideas.
c. Attempt some recommendations.
d. Possibly set up a committee of interested people.
If you have interest in the subject please plan to attend.
INTRODUCTION
One of the benefits eXPected from the use of Computer Networks is the
sharing of data among users of the system. This paper is an attempt
to classify the issues involved, discuss some approaches that might
be taken to achieve the goal of facilitating data sharing and to
point out some advantages and disadvantages of these approaches.
CONSIDERATIONS
In the process of selecting an approach one has to consider the
following issues:
1. Does the approach provide the use of one language to Access all
data on the network?
2. Does the approach facilitate sharing of existing data created
and manipulated by existing data management systems?
3. Does the approach encourage users to share data and use the
facility provided? How evolutionary is the approach?
4. Could a failure of one node in the network cause the failure of
the data sharing facility?
5. Does the approach promote or hinder further development of data
management systems?
6. What are the implementation considerations?
7. What are speed considerations?
POSSIBLE APPROACHES
1. Centralized data management system (CDMS).
This approach is consistent with the idea that a Computer
Network eventually will evolve into a collection of specialized
service nodes, where each node would perform a specific
function well. Users will use services on nodes according to
their needs. For example, one node could be a PL/I machine
(possibly a microprogrammed machine to perform PL/I compilation
efficiently), another node could be a "number cruncher" for
parallel-structured problems (ILLIAC IV), etc. In the same way
there will be a node responsible for all data management needs
for the network.
Depending on the assumptions made one of two ways can be
chosen:
a. As assumption that we must be able to share all data,
implies that the same data management system can create and
manipulate this data, and therefore must perform all the
functions required of a data management system, regardless
of the particular use. It is generally agreed that such a
task is monumental and impractical (if not impossible),
since different data management systems are designed to
perform specific functions well on the expense of degraded
performance of other functions (e.g., fast retrieval of
large files, limited updating capabilities).
b. The assumption is made that users will share only data from
the same file on a particular data management system. In
this case one can implement different data management
services for different tasks, but put them all on the same
node to provide a data management service to the Network
users. This approach can still use one common language to
access these services. This is apparently the approach
taken by CCA as indicated in NIC memo 5791.
2. Standardized data management system (SDMS).
In this approach a particular data management system is adopted
to be implemented on all nodes. This provides for a
standardized data management language as well as an identical
logical data structures. Alternatively, one can choose a set
of data management systems to be implemented on all nodes, then
be able to share information manipulated by the same data
management system on different nodes. This approach has many
drawbacks as will be discussed later.
3. Integrated data management system (IDMS).
This approach suggests the integration of local (to the node)
data management systems and local data (files) through the use
of appropriate interfaces and a common data management
language.
Under this category there may be different approaches depending
on the function of the interfaces:
a. There is an interface module in every node for every local
data management system. The interface performs a dual
function: on the way out--it issues requests in the common
language to remote nodes; on the way in--when a request in
the common language is received, the interface performs
translation from the common language to the local data
management language. From a single request the translation
might produce a series of commands in the local language
(for example, suppose that the local language permits the
specification of one quantifier only, such as "age<_41."
Suppose that the request received in the common language
specifies "list all names where age<_41 and children _>5."
The translation will produce a series of commands of the
form: "list all names where age <_41," "save the list
temporarily," "list all names in temporary file where
children>_5").
b. Move all local interfaces which were described above into
one central node. This node is now the service node. It
accepts a request in the common language and produces a
series of commands to all nodes involved, in their local
data management languages.
c. The local interface accepts the name of a local file (or
relevant portion of the file), and sends this file to the
requester after performing a translation of the data. The
data can be translated using a technique such as the "Form
Machine" (described in NIC 5772). The file is translated
from the local data management data structure to the
requesters data structure, so that the requester can perform
the desired function using his local data management system.
4. Unified data management system (UDMS).
This approach suggest the use of a standard interface which is
to be part of every data management system on the Network. The
interface has three ends. One to the user language, one to the
particular physical system used and one to the Network. The
interface should be global enough to permit separation of
system decisions from user language decisions. If this
interface is standardized on a Network, it will facilitate
communication between local data management systems in a
unified way, while permitting the development and evolvement of
different local data management systems. (This is a rough
description of the approach taken by Barry Wesseler in Utah.)
THE COMMON LANGUAGE
It is well known that the design of a language involves a compromise
between the ease of use of the language and its capability to express
the functions desired. A try to merge two languages usually results
in the worsening of one or both of these considerations.
For the purpose of having a common language for data management it
may be desirable to separate between the above mentioned
considerations. Use natural-language for ease of use, and a formal
intermediate language powerful enough to express any functions
desired. This is the approach taken in the development of CONVERSE
in SDC [1]. The intermediate language can be as complex as one likes
since it is invisible to the user.
DISCUSSION
Predictions for future use of computers (and therefore computer
networks) point out that "in 1975 we will process mostly data" [2].
Therefore, the problem of sharing data on a computer Network, as well
as accessing data from remote nodes in some common language are
extremely important.
If all that is desired is the sharing of data in a file by more than
one user, then the CDMS approach is appropriate. Approach la is
impractical, but lb can provide a valuable service. Selecting this
approach does not permit the sharing existing data which was created
with existing data management system, unless a restructuring of the
data for the CDMS is performed. This approach does not easily permit
the development of new data management systems since the CDMS should
stay stable for the Network use. It does not involve translation of
data or languages and therefore should provide good access speed.
The SDMS approach has many drawbacks. Selecting it implies the
imposition of a particular data management system on all nodes. It
inhibits further development. It does not permit the sharing of
existing information. The main advantage would be the modularized
structure so that the failure of one node cannot cause the failure of
the entire system. Also, because of the standardized approach
sharing of data from different nodes does not involve any
translation.
The main advantage of the IDMS approach is that it permits the
continued use of existing data management systems with existing data
bases associated with them while permitting the sharing of data among
the network community of users. Since it permits the continued use
of local data management systems it is the most evolutionary approach
and most likely to be accepted by a user of an existing data
management system. There are applications where users on each node
on the Network perform mostly local access of data, and less often
find it desirable to be able to share data with other nodes. For
example, if hospitals are connected to nodes of a Computer Network,
then most of the data about patients is accessed locally, but
sometimes it is necessary to access information from other hospitals,
such as global statistical information. The same situation exists
for criminal files, local branches of banks, credit bureaus,
warehouses, etc. Approach 3a permits the advantages of
modularization, but 3b is easier to implement since no additional
interfaces are necessary in the different nodes. Approach 3c seems
hard to implement and can introduce inefficiencies since it involves
translation from one data structure (which might be designed for
efficiency) to another data structure (which may not be as
sophisticated). It also involves the shipment of large amounts of
data across the network.
The UDMS approach permits the continued development of local systems
while facilitating a unified way for Network communication of data
requests. It is not clear at this point whether this approach is
practical.
Other important issues concerning sharing of data on a Computer
Network, and which are mentioned in [3] are overlap of information in
different files and the possibility of the same information to be
contradictory, security and privacy problems, sponsors of a file vs
users of a file, and others.
ACKNOWLEDGMENT
Discussions with the following people were very valuable: Al Vorhus,
Peggy Karp and others in MITRE, Barry Wesseler in Utah, Gerald
Levitt, N. Cohen and others in RAND, Clark Weissman, and Charlie
Kellogg in SDC, Richard Winter of CCA.
REFERENCES
1. Kellogg, C. "A Natural Language Compiler for Online Data
Management." Fall Joint Computer Conference Proceedings, Vol. 33,
part I, 1968. pp. 473-492
2. Clamons, Eric H. "Introductory Remarks to Data Base Management
Seminar." Proceedings of Workshop on Networks of Computers (NOC-
1969) NSA pp. 89-90
3. Hicken, George "Data Base Confrontation in an Information
Network." Proceedings of Workshop on Networks of Computers (NOC-
1969). NSA pp. 99-115.
[ This RFCwas put into machine readable form for entry ]
[ into the online RFCarchives by Ryan Kato 6/01]
