Processing Sequences: A new Paradigm for Out-of-Core Processing on Large Meshes

In this paper we introduce a new processing paradigm for meshes that are too large to fit entirely into main memory. We define the concept of a processing sequence that is essentially just a ordered sequence of triangles. This triangle ordering gives access to a mesh in a particular useful way that may be exploited to design efficient mesh processing algorithms on top of it. Rearranging mesh triangles in a particular order is already used for fast rendering on modern graphics cards with vertex cache. The triangles are sent to the card in a rendering sequence that tries to minimize cache misses. We exploit a similar strategy for more efficient mesh processing—but at a much larger scale. The main memory as a “cache” is much more flexible so that the data necessary for a complete mesh traversal can always be kept in-core. Therefore the analogue of a “cache miss” does fortunately not exist. In this working draft we define which triangle orderings constitute a processing sequence, discuss which other properties we want a processing sequence to have, sketch out different approaches for generating processing sequences, describe the two computational abstractions, namely boundary-based and buffer-based processing, provided by processing sequences, and give some simple examples of how to use our prototype API for out-of-core processing.