If you’re limited in RAM and can’t upsize, then yes, this does appear to be a good trade off. You can always refactor later and normalize if necessary. BTW, al

If you’re limited in RAM and can’t upsize, then yes, this does appear to be a good trade off. You can always refactor later and normalize if necessary. BTW, although it wouldn’t have helped for your specific benchmark schema creation of TYPES, I’ll plug my genSQL tool [0] for generating random data. It’s primarily designed around MySQL, but it can produce CSVs easily, which every DB can load. Turns out a lot of random() calls in most languages is slow af, so mine avoids that by (mostly) batching them in a C library. Should be able to create a million somethings in under 10 seconds on modern hardware in Python 3.11. [0]: https://github.com/stephanGarland/genSQL https://github.com/stephanGarland/genSQL