Development of the DoD protocol reference model

The Department of Defense has embarked on a Protocol Standardization Program aimed at the development of high-level protocol standards for the DoD internetwork environment. A major piece of this standardization program involves the development of the "DoD Protocol Reference Model" [1], which will serve to describe a basic framework for future DoD protocol standardization activities.

The DoD Reference Model is in many respects similar to the ISO Reference Model [2]; in other respects, there are major differences between the two models. This paper discusses the issues faced during the development of the current draft of the DoD Reference Model.

There are four fundamental differences between the DoD and ISO Reference Models: (1) the meaning of "layer"; (2) the importance of internetworking; (3) the utility of connectionless services; and (4) the approach to management functions.

It has been found that well-designed computer networks have their protocols organized hierarchically. By this we mean that one protocol interpreter provides a communications service to its users by adding value to the services provided by one or more other protocols; and "good engineering practice" dictates that at no point should one protocol be implicitly using its own services, i.e. the protocols form a hierarchy. Within the ISO Model, however, this requirement for protocol hierarchies is strengthened into a requirement that protocols fit into "layers". This is the origin of a major disagreement between the ISO and DoD Reference Models. Within the ISO Model, the concept of layer has become of paramount importance, overshadowing the more fundamental notion of protocol hierarchy. Good protocol engineering does not require that protocols fit into layers; it is only required that protocols be arranged hierarchically. This latter principle is the basis of the DoD Reference Model.

The second fundamental difference between the ISO and DoD Models is the relative importance given to the problems of internetworking. This is most easily seen by considering that the DoD Reference Model has explicitly identified an "Internet Layer". Although many efforts based on the ISO work are now taking the Network Layer to include an Internet Sublayer, this seems to confuse rather than clarify the issue (since it raises questions as to the meaning of sublayers).

It is also difficult to correctly present the DoD approach to internetting within the confines of the ISO Model due to the ISO Model's lack of emphasis on "connectionless" services. This is the third of the fundamental differences between the two models. The primary use of connectionless service within the DoD architecture is for internetting - IP [3] being the paradigm. However, connectionless services are of great utility in other contexts as well. For example, interactions with a Name Server may involve the use of a connectionless protocol within the Transport Layer.

The final fundamental difference between the ISO and the DoD Models is the manner in which various management - related functions are treated. Here we refer to functions such as the naming of resources, the control of access to resources, and the accounting for resource and network usage. Many of these functions are most naturally handled via connectionless services, they typically involve entities which are only involved at the initiation or termination of a sequence of data transfers, and many communicants may be involved besides the "users". The DoD Model identifies such protocols as typical of " session" services. Such protocols are placed within the session layer by virtue of their use of " transport layer" services and their common features.

Of course, there will be many protocols defined which sit at other locations within the protocol architecture which may play a role in the management of the network and of network resources. It is not implied that all such protocols are most appropriately defined as "session" protocols.

These are the differences between the two developing Models which are currently identified. Both Models will continue to evolve, most likely in directions which differ in important respects.