Distributed Ray Tracing of Large Scenes using Spark

We present work on implementing a ray tracing system that can work with distributed geometry so that scenes can exceed the memory requirements of a single machine using the Apache Spark big-data system. We look at a progression of implementations beginning with implementations that simply distribute the work and moving to implementations that allow the geometry itself to be distributed. We find that Spark makes it rather simple to distribute work across a cluster. While distributing the geometry across the cluster is more challenging, the implementation is still rather short given that it is capable of rendering scenes that are larger than any of the individual machines in the cluster could handle.