Next: , Previous: Introduction to distributed systems, Up: Introduction to distributed systems


1.1.1 Using OS Network Services

There are several programming techniques for developing distributed applications. These applications have traditionally been developed using network programming interfaces such as sockets. Programmers have to perform explicit calls to operating system services, a task that can be tedious and error-prone. This includes initializing socket connections and determining peer location, marshalling and unmarshalling data structures, sending and receiving messages, debugging and testing several programs at the same time, and porting the application to several platforms to uncover subtle differences between various network interfaces.

Of course, this communication code can be encapsulated in wrappers to reduce its complexity, but it is clear that most of it can be automatically generated. Message passing diverts developer's attention from the application domain. The query and reply scenario is a classical scheme in distributed applications; using message passing for such a scheme can be compared to only using the “goto” mechanism in a non-distributed application. This is considered unacceptable methodology in modern software engineering. A cleaner and more structured approach consists in using subprograms.

In some respects, network programming can be compared to parallel programming. The user can decide to split his code into several pieces and to multiplex the execution of threads himself, using a table-driven model. The scheduling code ends up embedded in the user code. This solution is error-prone and fragile in regard to any future modification. Relying on an implementation of threads such as provided in a POSIX operating environment is a better solution. Relying on language primitives that support concurrency, such as Ada tasks, is best, as the underlying parallelism support is thus entirely abstracted.