| Wireless mobile access to the Internet is an indispensable tool for private and corporate users alike. Research on in-stream clipping servers focuses on exploiting inherent parallelism on a variety of levels, but seems to have missed one that this research uncovered. The syntax of the markup language that describes a typical Web page has a tree-like structure, which suggests inherent parallelism, provided the data in one branch of the syntax tree can be processed independently of the others. This research examines the potential for improved quality of service to a handheld device on a wireless link by using a parallel cluster computing architecture as an in-stream proxy server, exploiting parallelism during the markup language transcoding process.; Comparing a parallel computing architecture's performance to that of a uniprocessor can be misleading if parallel software is simply reconfigured to run on a single machine, because the parallel management overhead adds delay to the single system's response time. This research compared each possible configuration of the distributed system's performance against that of a single system running equivalent software that was designed for a single system environment.; The experimental test bed consisted of a wireless mobile device simulator, a Web server, and a cluster server running the clipping service. The distributed version of the clipping server had from one to six active worker nodes, each with from one to three transcoding processes, for eighteen unique configurations. Experiments measured average response times for both the single-query and query-stream cases, and tested the scalability and robustness of both systems.; Distributed clipping performance was found to be significantly better than that of the single system, and the overhead was considerably less than that of similar systems reported in the literature, confirming initial expectations. Stress testing showed the single system to be more robust than the distributed version, however, by a factor of two. The distributed system's need for significantly greater network resources led to saturation and premature failure under stressful loads. |
