Rendering

Comparison between cluster and single machine rendering
. . . Reliability as the number of nodes increases
. . . Cost of licensing for commercial software
. . . Machine differences in heterogeneous clusters

Rendering is typically considered trivially parallel
Distribution methods, within frame and between frame (eg: animations)
Load averaging, within frame and between frame
. . . Variable size image segmentation, based on preview images
. . . Dynamic job distribution based upon load
. . . Using a non dedicated cluster, being a good farm citizen

Scientific data and large datasets
Example: Mars topology rendering (8 million polygons)

Light simulation vs raytracing/radiosity
. . . Importance for interior lighting, Architectural rendering
. . . Realism through physics
. . . Area lights, source distributions, physically measurable materials
. . . Example: Radiance

Bandwidth, more processors doesn't necessarily reduce rendering time!

Realtime rendering, the Holy Grail
. . . Interactive with simpler rendering -> Offline with increasing realism
. . . 60 frames per second -> 100 hours per frame!

Rendering on the Swinburne Farm
The Swinburne Supercomputer farm is available for approved
rendering projects both for student and commercial projects.