Loading...
Please wait, while we are loading the content...
Similar Documents
Hybrid Scheduling for Parallel Ray Tracing
| Content Provider | Semantic Scholar |
|---|---|
| Author | Katwijk, Jan Van |
| Copyright Year | 1996 |
| Abstract | Preface Ray tracing is an easy algorithm to explain, but difficult to implement efficiently. And even then performance is a difficult issue, which calls for every sequential and parallel optimisation that one can think of. This report relates a number of design choices which allowed a sequential ray tracer to evolve into a system for parallel ray tracing and is the result of a two-year (TWAIO) course on parallel rendering, undertaken by the author at the As ray tracing is one of those applications that do not have a regular structure, such as for example solving large matrices and sorting algorithms etc., good parallel solutions of ray tracing are likely to exhibit rather complicated control and data flow. It is far from easy to predict which objects will be referenced most and which processor is going to need what data in what time step. Regular problems are easy in this sense, so that, dependent on the type of application, either data parallel solutions or demand driven approaches usually result in good speed-ups. Because ray tracing and radiosity algorithms lack sufficient structure, for each task generated it should be decided whether it is cheaper to transfer the task to the processor that stores the relevant data, or whether the appropriate data should be fetched from a remote processor. This basic idea leads rather naturally to a combination of data parallel and demand driven scheduling, called hybrid scheduling. Coherence can be used to distinguish between demand driven tasks and data parallel tasks. If there is a number of tasks that will reference the same small amount of data, these tasks can easily be sent to a processor along with the data in demand driven manner. If for some set of tasks the amount of data referenced becomes too large, then these tasks are better executed data parallel by sending the tasks to the processors that already store the data. This scheme was suggested by Jansen et. al. [11] to be applied to parallel ray tracing. To prove the validity of this assumption was the main topic of this project. The TWAIO design engineering course consists of two parts, each of which takes roughly one year to complete. The first year was filled with courses on parallel rendering and computer graphics, which created a basic understanding of the area of research. The second year of the course was devoted to the TWAIO engineering … |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.cs.bris.ac.uk/Publications/Papers/1000155.pdf |
| Alternate Webpage(s) | http://www.cs.bris.ac.uk/Tools/Reports/Ps/1996-reinhard-1.pdf |
| Alternate Webpage(s) | http://www.cs.bris.ac.uk/~reinhard/papers/1996-reinhard-1.pdf |
| Alternate Webpage(s) | http://www.cs.utah.edu/~reinhard/papers/1996-reinhard-1.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
