HORT: a High-Performance MapReduce Framework for Ray Tracing in the Cloud

Lesley Northam, Rob Smits, and Khuzaima Daudjee

Ray tracing is a type of high-quality computergenerated imagery rendering that makes heavy demands on computing resources. We present a new high-performance ray tracing framework for cloud computing, allowing computing resources to be dynamically provisioned within an existing infrastructure on a pay-per-use basis. Our framework, called HORT, utilizes the MapReduce programming paradigm, which subdivides tasks that can be computed in parallel. HORT is easily scaled to any cloud infrastructure-as-a-service configuration while providing faulttolerance intrinsically. Importantly, our framework is effective and efficient in handling large scenes by eliminating the performance overheads associated with replicating data and on-demand data requests that are required of other ray tracing frameworks that employ distribution and parallelism. We demonstrate that HORT can also utilize efficiently the parallelism of GPUs with MapReduce in the cloud to offer high-performance.


Submission in progress.
A SIGGRAPH poster regarding HORT was presented in 2011.


For more information regarding HORT, including the current status or results, please contact me directly.