Design And Implementation Of Network Programming Framework For Distributed Computing

Cloud computing is a central part of the third IT revolution, and distributed computing plays a critical role in it. The establishment of a distributed system is a mainstream way to implement distributed computing, which uses ordinary commercial hardware and achieves expansion by increasing the number of machines. A Distributed system is hard to implement, because it involves multiple processes of multiple hosts, and the process between different hosts communicated by network. These complexity can eliminate by using object-oriented design techniques and features of high-level programming language. Network and Distributed System Laboratory’s foundation class library was originally designed to eliminate the complexity of the implementation of communication servers. The main objective of this thesis is designing and implementation a network programming framework for distributed computing based on foundation class library.This thesis analyzes the requirements of the implementation of distributed computing business, and then refine two models: communication models and concurrency models. We separate the functional code from the non-functional code by separating communication models and concurrency models from the business logic. According to these two models, this thesis design and implement two framework for supporting the implementation of distributed computing business: the multi-threading framework and the event-driven framework. The multi-threading framework provides a very natural abstraction for high concurrent business, and also has good business isolation, but threads are scarce resources. The event-driven framework provides a lot of concurrent and also has no context switch and no lock, but it increases programming complexity. Combining with the practical applications, this thesis presents a programming paradigm: the main process implement on the event-driven framework, uniformly handle network I/O events, timer events and signal events, while the multi-threading framework uniformly handle any other blocking process, such as disk I/O.Finally, this thesis validates the functional and performance characteristics by using test cases. It not only can satisfy the performance requirements, but also greatly simplify the workload of development by using network programming framework when implement distributed computing, and enable programmer focus on business logic.