Dynamic Chunking for Out-of-Core Volume Visualization Applications

Given the size of today’s data, out-of-core visualization techniques are increasingly important in many domains of scientific research. In earlier work a technique called dynamic chunking [1] was proposed that can provide significant performance improvements for an out-of-core, arbitrary direction slicer application. In this work we validate dynamic chunking for several common data access patterns used in volume visualization applications. We propose optimizations that take advantage of extra knowledge about how data is accessed or knowledge about the behavior of previous iterations and can significantly improve performance. We present experimental results that show that dynamic chunking has performance close to regular chunking but has the added advantage that no reorganization of data is required. Dynamic chunking with the proposed optimizations can be significantly faster on average than chunking for certain common data access patterns.