RE: Transforming large XML docs in small amounts of m

> It's purely XSLT 1.0, using Saxon (on Linux and Windows, if 
> that matters...), although suggestions to change this would 
> not be shunned.

Just as an experiment, try it on Saxon 8.x rather than Saxon 6.5. It doesn't
always make any difference, but on some occasions I've seen it make a
dramatic improvement.

> I think that transforming 150Mb of data in 400Mb 
> of RAM would be a sensible target (is this sensible?)

That's ambitious. To achieve that, you're going to have to do something that
condenses the input document before transformation.

> I'm not sure how to tell what proportion of the memory is 
> used for the input DOM, output DOM, etc...

If you really mean "DOM", then start by not using a DOM, and using a Saxon
TinyTree instead. (Saxon doesn't use any memory for the final result tree -
it's normally piped straight into the serializer).

Michael Kay
http://www.saxonica.com/