Network Working Group R. Watson
Request for Comments: 592 SRI
NIC 20391 November 1973 Some Thoughts on System Design to Facilitate Resource Sharing
There is a growing interest in moving toward more resource sharing on
the ARPANET. Some resource sharing has been taking place by having
systems open TELNET connections and generating user command strings.
I think that this is fine for experimental use, but is not the way we
want to operate in real usage. What I believe network system
builders should do is to develop mechanisms appropriately designed
for computer-computer communication.
SYSTEM INTERCONNECTION, AN APPROACH
The goal I would like to see us move toward is to view all systems on
the network as offering certain service modules, any subset of which
can be combined in building other systems. Each service module would
have a well advertised set of primitive service capabilities that it
could provide. It would have documented commands at the level of
present Telnet or FTP commands for gaining access to its services.
It would also have a defined network connection procedure. Then any
system builder wanting to avail himself of these services could do so
and integrate them into his own user interface environment.
At the present time when a system is built, the system builders tend
to see it as a stand alone thing or at most something to be used
within a specific environment. What I would like to see fostered is
the idea that any system built is not only a stand alone environment
but also a network service or set of services. The builders would
define not only a user interface for their environment, but also a
set of primitives and primitive commands that can be accessed by
other systems around the network to get that service performed.
For example, we are redesigning the NLS Journal in light of our
experience and that of Network Mail as a set of protocols and
services. If one looks at the processes of the NLS Journal one
can see a number of separate services that could be provided by
different network sites or combined in varying combinations by a
single site. These being:
Distribution (identification of addressees and maintainance of
the required data bases being a related service), recording
(numbering and storing of items), cataloging, and retrieval.
At the moment these services are fairly tightly interconnected in the
NLS Journal and what we want to do is to decouple them and define
their intercommunication by protocols that would allow them to be
distributed in different hosts on the network. Mechanisms would also
be defined for the several hosts performing similar services around
the network to work together cooperatively.
As a further example, there are also other services that NLS could
probably provide such as structured file creation and manipulation;
information portrayal online or in hardcopy; database querying etc.
However, at the moment the system is not explicitly structured from
the point of view that outside systems could come into it anywhere
but at the human user interface even though internally it is quite
modular. It would be straightforward for us to identify those NLS
services that other system builders might possibly be interested in
incorporating into their systems with their own user interface and
then to do the restructuring and primitive command definition
necessary. Other groups building systems on the network could
perform a similar examination.
CCA, on the other hand as I understand it, has taken this point of
view from the beginning, namely building the Datacomputer on the
assumption that it is primarily a network resource and is to be used
by other systems. BBN is also moving in this direction in the design
of Distributed TENEX.
There is nothing new in the above ideas; they come from generalizing
past successes we have all had with network protocol development and
with good software engineering practices. It will, however, take a
change in the thinking of system designers, some concrete examples,
and ongoing dialog to make such a design philosophy the normal
network way of life.
SOME FUNCTIONS READY FOR INTERCONNECTION
The area of dialog support may be the first area ripe to create such
a synthesis with the several systems in or coming into existence,
each solves part of the problem (with some overlap). The dialog
support systems on the network known to me are:
The NLS Journal (supports recorded and cataloged dialog and linked
networks of documents and messages).
NLS Screen linking and splitting (supports close collaboration of
two or more people working together in real time in NLS)
The network wide linking of terminals through BBN's RSEXEC.
Tenex Sndmsg and Readmail and other mail systems support
nonrecorded dialog and further manipulation of received messages.
(Some interconnection between NLS and these facilities has been
The communication system under design at USC-ISI to support a
range of message services.
The online conferencing system being built by Jim Calvin of Case,
John Iseli of Mitre and others supports online conferencing of
several members and has facilities to utilize various Tenex
subsystems such as TECO and NLS to support conferees.
The Hack system of CASE offers a bulletin board service.
The Forum system of IFF supports online and distributed in time
conferencing and other features.
Other areas possibly ripe for synthesis are 1) file and data
management, and information retrieval services; 2) editing and
hardcopy portrayal with systems like Tenex RUNOFF, SU-AI's PUB and
SRI-ARC's Output Processor.
If the salient service features, concepts, goals of each could be
defined clearly and appropriate service primitives, as per other
ARPANET protocols, could be defined for each, anyone wishing to
incorporate that service with a user interface appropriate to his
environment or philosophy could do so.
SYSTEM INTERCONNECTION ISSUES RELATED TO THE ABOVE PROPOSAL
There are many detailed issues related to system interconnection as
proposed above. A number seem worth mentioning here.
1) Types of Network Connections
The number and type of network connections to be opened between
classes of cooperating processes can probably be systematized.
One of the important elements of the FTP and Graphics protocol
efforts was to define the number and type of connections necessary
for these classes of transaction. Similar classification and
connection definition will be required for other types of
2) Data Structure Translation
The whole area of translation and transfer of data structures more
complicated than sequential files needs vigorous thought and
Systems built around sequential files are presently dominant on
the ARPANET and provide a base for simple useful economical
tools. I, however, do not believe that the longer run tool
sharing can depend on communication between sequential files,
but requires structured files. Experience with NLS tree
structured files shows that even this level of structuring may
be inadequate for many uses and more sophistication may be
required. A similar trend exists in work with computer
graphics and generalized data management systems. Developing
protocols for handling structured data bases or agreement on
common structuring characteristics seems an important need.
Factors influencing responsiveness to users in an environment of
heavy geographically separated resource sharing need determination
4) Documentation of System Interfaces
It is probably reasonably straightforward to define service
interfaces, but they will be useless unless their activating
command languages and other conventions are well documented and
this documentation is kept up to date.
A very difficult problem once you interconnect systems at lower
levels is to design an appropriate network accounting and banking
system that will not cause undue delays in accessing distributed
6) Error Handling
We need to develop mechanisms for passing error signals around
when system environments are crossing machine boundaries.
7) Standard Parameter Formats
Data types such as strings, integers, floating point numbers,
arrays, pointers, etc. need to have standard representations
defined for passing parameters back and forth between machines.
8) HELP at the Procedure Call Level
A HELP mechanism needs to be defined in the protocols to provide
information that each designer can translate to his user
interface. Standards for requesting HELP information and
structuring HELP data bases needs agreement.
I wish to acknowledge the useful suggestions of Charles Irby and Jim
White in the thoughts above.