M.Sc. Tezi Görüntüleme

Student:	Mehmet Bilsay KARADENİZ
Supervisor:	Asst. Prof. Dr. Cemal KÖSE
Department:	Computer Engineering
Institution:	Graduate School of Natural and Applied Sciences
University:	Karadeniz Technical University, Turkey

Title of the Thesis:	Time-Varying Volume Data Visualization on Distributed Systems
Level:	M.Sc.
Acceptance Date:	14/7/2006
Number of Pages:	97
Registration Number:	i1707

Summary:

Time-varying volume data visualization is a computationally intensive process and requires large amount of memory. Parallel processing offers the potential for achieving the visualization in reasonable times.

This thesis discusses the potentials offered by small computer clusters in parallel visualization of time-varying volume data.

Two load balancing approaches are implemented for an efficient parallel splatting. This parallel splatting approach exploits time and object space-slicing technique. Here, MPI is used for message passing interface. This standard message-passing interface makes it easy for developers to concentrate on parallel programming, not to low level communication details.

The Mandelbrot set is used as time-varying volume data in visualization. This 4D Mandelbrot sets are calculated in the quaternion.

By employing our approach, the effects of the number of computers, processes per computer, volume slices and voxels, Gauss mask size, surface normal extraction volume size, and load balancing algorithms on the performance are examined and the results are evaluated.

As a result of this study it is seen that; especially static load distribution method has negative effects on the performance for unbalanced loads. In addition to this, as the problem size increases the efficiency of the system decreases. Also the optimum number of processes per computers is found to be two.

Keywords: Volume Rendering, Parallel Splatting, Time-Varying Volume Data, Load Balancing, Distributed Systems, Mandelbrot Set.