2001.04.26 16:13 "RE: 8K chunks for RowsPerStrip?", by Max Martinez

That's funny, because the ONLY single strip images I have ever seen are 1-bit, where it is very common to only have 1-strip.

Perhaps there needs to be an inflationary provision in the recommended maximum strip/tile size but to abandon it is to partially abandon the promise of interoperability attempting to be delivered through the TIFF format.

A single strip of a full Landsat TM Scene (~350 MB) is absurd even today. As far as I know, despite the increase in RAM and disk space for swap, no one has yet upped the present 2GB virtual address space limit imposed in most operating environments. Thus, one might conclude that the 8K recommendation should stay right where it is because, in the old space/time battle, disk access times and cpu times have gotten faster while virtual address space has remained constant. You are incurring less of a penalty in terms of time to access multiple strips than you were when they wrote the spec, but the relative space cost for having strips from multiple files reside simultaneously somewhere in the virtual address space of a process has remained the same.