William R. Otte and Douglas C. Schmidt
Developing software for networked applications is hard, and developing reusable software for networked applications is even harder. First, there are the complexities inherent to distributed systems, such as optimally mapping application services onto hardware nodes, synchronizing service initialization, and ensuring availability while masking partial failures. These complexities can stymie even experienced software developers because they arise from fundamental challenges in the domain of network programming.
Unfortunately, developers must also master accidental complexities, such as low-level and nonportable programming interfaces and the use of function-oriented design techniques that require tedious and error-prone revisions as requirements and/or platforms evolve. These complexities arise largely from limitations with the software tools and techniques applied historically by developers of networked software.
Despite the use of object-oriented technologies in many domains, such as graphical user interfaces and productivity tools, much networked software still uses C-level operating system (OS) application programmatic interfaces (APIs), such as the Unix socket API or the Windows threading API. Many accidental complexities of networked programming stem from the use of these C-level OS APIs, which are not type-safe, often not reentrant, and not portable ...