Performance of CORBA-based parallel architectures.

摘要

Distributed Computing Systems have shown advantages over centralized systems with regard to reliability, scalability, extensibility, high performance and cost effectiveness. However, the heterogeneous environment inherent in Distributed Computing Systems is a major challenge for software developers. With wireless data computing, this challenge is even greater, due to the variety of mobile devices, and differing access technologies and communication protocols. Middleware provides a good solution for distributed systems in terms of interoperability, compatibility, and easy application development in wired and wireless networks. However, middleware often adds system overheads that can be detrimental to system performance.; To satisfy the performance requirements of wired and wireless data computing, three new CORBA-based parallel client-server architectures are investigated in this thesis: the Handle-driven Process Planner Architecture (HPP), the Forwarding Process Planner Architecture (FPP), and the Synchronous Process Planner Architecture (SPP). This thesis investigates the relative performances of these three architectures, as well as the traditional sequential architecture in wired and wireless environments. This research thoroughly analyzes the impact of multithreading, network bandwidth, message size, server demands, and the wireless link packet drop rate on the latency and scalability attributes of the three architectures.