Comment by jklowden
8 hours ago
Nonsense. See Codd’s first paper.
1NF removes repeating groups, putting for example data for each month in its own row, not an array of 12 months in 1 row.
Storage efficiency was never the point. IMS had that locked down. Succinctness of expression and accuracy of results was the point. And is: normalization prevents anomalous results.
To expand on this point: unexplained "distinct" is often a code smell
Normalizing repeating groups doesn't offer significant savings when they are completely populated (e.g. each entity has the full 12 monthly values per year), but other types of normalization do. For example dependent data are actually redundant.
I think parent was saying it’s a benefit, not the original purpose. If I store a FK to a table containing my company’s corporate address, that is a tremendous savings in storage (and memory pressure), and it also eliminates update anomalies.
And other people are saying it may not always have that effect.
Normalization ultimately boils down to breaking your data down into the most elemental and simplest "facts" about them, and its greatest value is how it allows and encourages more flexible and disparate ways of looking at the same information.