Previous: Using a Middleware Environment, Up: Introduction to distributed systems


1.1.3 Using a Distributed Language

Rather than defining a new language like the CORBA IDL, an alternative is to extend an existing programming language with distributed features. The distributed object paradigm provides a more object-oriented approach to programming distributed systems. The notion of a distributed object is an extension to the abstract data type that allows the services provided in the type interface to be called independently of where the actual service is executed. When combined with object-oriented features such as inheritance and polymorphism, distributed objects offer a more dynamic and structured computational environment for distributed applications.

The Distributed Systems Annex (DSA) of Ada defines several extensions that allow the user to write a distributed system entirely in Ada. The types of distributed objects, the services they provide, and the bodies of the remote methods to be executed are all defined in conventional Ada packages. The Ada model is analogous to the Java/RMI model. In both languages, the IDL is replaced by well-defined language constructs. Therefore, the language supports both remote procedure calls and remote object method invocations transparently, and the semantics of distribution are consistent with the rest of the language.

A program written in such a language is intended to communicate with a program written in the same language, but this apparent restriction has several useful consequences. The language can provide more powerful features because it is not constrained by the common features available in all host languages. In Ada, the user will define a specification of remote services and implement them exactly as he would for ordinary, non-distributed services. His Ada environment will compile them to produce a stub file (on the caller side) and a skeleton file that automatically includes the body of the services (on the receiver side). Creating objects, obtaining or registering object references or adapting the object skeleton to the user object implementation are made transparent because the language environment has a full control over the development process.

Comparing with multi-threaded programming once again, the language extension solution is equivalent to the solution adopted for tasking facilities in Ada. Writing a distributed application is as simple as writing a concurrent application: there is no binding consideration and no code to wrap. The language and its run-time system take care of most issues that would divert the programmer's attention from the application domain.