Master's Thesis Al-waleed Shihadeh

Building a 3D web collaboration environment is a huge challenge as it involves the development of a high performance web application which allows engineers to view and exchange 3D models over the Internet. This is considered to be a huge challenge because the current status of the 3D Internet is still evolving, as well as the complexity of loading large 3D files into the browsers. Fortunately, this objective can be achieved by using current 3D technologies, like X3DOM and X3D. However, when working with large 3D models, it is not enough to just rely on the performance of these technologies to build a high performance application.

Large 3D models consume a large amount of memory and therefore, rendering these models will take a long time. One way to enhance the performance of rendering 3D models is to transcode the geometries of these models to binary files. In addition, the transcoding process will enhance the interactions between the 3D models and the users since the size of these models is much smaller. To minimize the required transcoding time, a data-parallel transcoding approach should be used. The aim of this thesis is to investigate the possibility of developing a data-parallel transcoding system based on the Hadoop framework. The outcome of this thesis is a web data-parallel transcoding system that provides the end users with a 3D data-parallel transcoding service and a 3D model viewer.

The implemented system is a distributed application that allows engineers to upload a single X3D document, transcode it to binary format and deploy it on the web. In addition, this thesis provides an evaluation and performance analysis of the implemented transcoding approach. The main outcome of the evaluation part is that using a data-parallel approach for transcoding 3D data will dramatically reduce the required transcoding time compared to sequential transcoding approaches. In addition, the
main properties that influence the required transcoding time are the shapes number, files number, split size and cluster nodes number.