Currently the way data is passed to the TERR engine from Spotfire is through a file saved to disk and then read by TERR.
This is very time-consuming and inefficient when using millions of rows with many columns.
For TERR to be useful in this case it would require that the data already in memory could be passed by reference (maybe copy-on-write) to TERR so that speed would be comparable to e.g. filters and aggregate functions native to Spotfire.