We present a combination of techniques for interactive out-of-core
isosurface extraction and rendering of time-dependent data sets. We
make use of an index tree that allows extraction of all
isovalue-spanning cells from any time step, and for any isovalue, at
rates of several hundreds of frames per second. This data structure is
constructed in a pre-processing stage, and effectively uses the
temporal coherence in the data set. However, for very large data sets
such as those resulting from CFD simulations, this tree structure can
easily become too large to fit in main memory. Therefore, we have
adapted the data structure for out-of-core application by adding an
intelligent paging scheme, which allows interactive exploration of
very large data sets on a normal PC. Only a user-specified time window
will be kept in main memory and other parts of the tree will be read
and released on-demand. This paging scheme was implemented using
multi-threading. Finally, to avoid time-consuming triangulation and
surface reconstruction, we have used a hardware-assisted direct point
rendering algorithm, achieving interactive rendering frame rates.