Developing a System Engineering View

Stakeholder and Concerns      Modeling the View      Key Issues     Introduction    Client/Server   Master/Slave     Peer-to-Peer    Distributed Object Management


Stakeholder and Concerns

This view should be developed for the systems engineering personnel of the system, and should focus on how the system is implemented from the perspective of hardware/software and networking.

Systems engineers are typically concerned with location, modifiability, reusability and availability of all components of the system. The System Engineering View presents a number of different ways in which software and hardware components can be assembled into a working system. To a great extent the choice of model determines the properties of the final system.  It looks at technology which already exists in the organization, and what is available currently or in the near future. This reveals areas where new technology can contribute to the function or efficiency of the new architecture, and how different types of processing platform can support different parts of the overall system.

Major concerns for this view are understanding the system requirements. In general these stakeholders are concerned with assuring that the appropriate components are developed and deployed within the system in an optimal manner.

Developing this view assists in the selection of the best configurations for the system.

Modeling the View

The architecture of the overall hardware / software system can be modeled with ADML (components, ports, connectors, and roles).

Key Issues

This view of the architecture focuses on computing models that are appropriate for a distributed computing environment. To support the migration of legacy systems, this section also presents models that are appropriate for a centralized environment. The definitions of many of the computing models (e.g., host-based, master-slave, and three-tiered) historically preceded the definition of the client/server model, which attempts to be a general-purpose model. In most cases the models have not been redefined in the computing literature in terms of contrasts with the client/server model. Therefore, some of the distinctions of features are not always clean. In general, however, the models are distinguished by the allocation of functions for an information system application to various components (e.g., terminals, computer platforms). These functions that make up an information system application are presentation, application function, and data management.

 

Client/Server Model

Figure 1: Basic Client/Server Model

Client/server processing is a special type of distributed computing termed co-operative processing because the clients and servers co-operate in the processing of a total application (presentation, functional processing, data management). In the model, clients are processes that request services, and servers are processes that provide services. Clients and servers can be located on the same processor, different multiprocessor nodes, or on separate processors at remote locations. The client typically initiates communications with the server. The server typically does not initiate a request with a client. A server may support many clients and may act as a client to another server. Figure 1 depicts a basic client/server model, which emphasizes the request-reply relationships. Figure 2 shows the same model drawn following The Open Group Technical Reference Model, showing how the various entities and interfaces can be used to support a client/server model, whether the server is local or remote to the client. In these representations, the request-reply relationships would be defined in the API.

Figure 2: Reference Model Representation of Client/Server Model

Clients tend to be generalized and can run on one of many nodes. Servers tend to be specialized and run on a few nodes. Clients are typically implemented as a call to a routine. Servers are typically implemented as a continuous process waiting for service requests (from clients). Many client/server implementations involve remote communications across a network. However, nothing in the client/server model dictates remote communications, and the physical location of clients is usually transparent to the server. The communication between a client and a server may involve a local communication between two independent processes on the same machine.

An application program can be considered to consist of three parts:

In general, each of these can be assigned to either a client or server application, making appropriate use of platform services. This assignment defines a specific client/server configuration.

Master/Slave and Hierarchic Models

In this model, slave computers are attached to a master computer. In terms of distribution, the master/slave model is one step up from the host-based model. Distribution is provided in one direction-from the master to the slaves. The slave computers perform application processing only when directed to by the master computer. In addition, slave processors can perform limited local processing, such as editing, function key processing, and field validation. A typical configuration might be a mainframe as the master with personal computers (PCs) as the slaves acting as intelligent terminals, as illustrated in Figure 3.

The hierarchic model is an extension of the master-slave model with more distribution capabilities. In this approach, the top layer is usually a powerful mainframe, which acts as a server to the second tier. The second layer consists of LAN servers and clients to the first layer as well as servers to the third layer. The third layer consists of PCs and workstations. This model has been described as adding true distributed processing to the master-slave model. Figure 3 shows an example hierarchic model in the third configuration, and below, Figure 4 shows the hierarchic model represented in terms of the entities and interfaces of the Technical Reference Model.

Figure 3: Host-Based, Master-Slave, and Hierarchic Models

 

Figure 4: Hierarchic Model Using the Reference Model

 

Peer-to-Peer Model

In the peer-to-peer model there are co-ordinating processes. All of the computers are servers in that they can receive requests for services and respond to them; and all of the computers are clients in that they can send requests for services to other computers. In current implementations, there often are redundant functions on the participating platforms.

Attempts have been made to implement the model for distributed heterogeneous (or federated) database systems. This model could be considered a special case of the client/server model, in which all platforms are both servers and clients. Figure 5 (A) shows an example peer-to-peer configuration in which all platforms have complete functions.

Figure 5: Peer-to-Peer and Distributed Object Management Models

Distributed Object Management Model

In this model the remote procedure calls typically used for communication in the client/server and other distributed processing models are replaced by messages sent to objects. The services provided by systems on a network are treated as objects. A requester need not know the details of how the object is configured. The approach requires:

This approach does not contrast with client/server or peer-to-peer models but specifies a consistent interface for communicating between co-operating platforms. It is considered by some as an implementation approach for client/server and peer-to-peer models. Figure 5 presents two distributed object model examples. Example B shows how a client/server configuration would be altered to accommodate the distributed object management model. Example C shows how a peer-to-peer model would be altered to accomplish distributed object management.

The Object Management Group (OMG), a consortium of industry participants working toward object standards, has developed an architecture - the Common Object Request Broker Architecture (CORBA), which specifies the protocol a client application must use to communicate with an Object Request Broker (ORB), which provides services. The ORB specifies how objects can transparently make requests and receive responses. In addition, Microsoft's Object Linking and Embedding (OLE) standard for Windows is an example of an implementation of distributed object management, whereby any OLE-compatible application can work with data from any other OLE-compatible application.


Copyright The Open Group, 1998, 2000