> Tuple table performance, though, tends to degrade quickly when querying separate rows for every little attribute.
Interesting. Do you know what causes this steep decline in performance?
A larger sequential read is typically not that much slower regardless on what sort of drive you are using. I'd have expected that the random access pattern of a normalized table would lose out.
More _memory_ and row caching in RAM has allowed for a resurgence in denormalized schema.
FB's underlying schema is similarly just a few tables. But the entire dataset is memcached.