I'm surprised they didn't select any tasks that couldn't easily fit on a single machine (largest input set was <200GB). That said, if the scale up performance c

I'm surprised they didn't select any tasks that couldn't easily fit on a single machine (largest input set was <200GB). That said, if the scale up performance can be improved without compromising scale out, why not?