Well it is indeed a tough nut. I just got access to UDN, there I was able to get into more details. Still, if we come up with a solution I’ll update this question.
And yes, we’ll definitely reduce the amount of data and derive whatever we can on the fly, even if the result is not 100% precise.