Source: Facebook

OVERSING UPDATE

I think were we are at the point on Oversing that I get to say “I told you so” about denormalization.

I love php because of its use of strings. you can store all sorts of nonsense like they do in WordPress, and just pull out what you need, rather than building lots of dimensions. For entities with limited relations it’s just so wonderful. And if you need more information you denormalize to a document database (non-relational) for quick retrieval.

Oversing is pretty demanding. But we benefit from the fact that unlike say, Facebook, and more like say Atlassian, we don’t commingle customer data. So that means that the largest database we encounter is a single large customer.

This one issue we have left (we tried one experiment and it doesn’t work) is one in which the workspace tells you a lot about the counts of each filter you can apply. And in normalized form these are individually inexpensive but collectively costly.

So we sort of have this choice: Make everything visible but use security only to limit access (most likely). Denormalize everything we can and hope that the tests hold up. (increases fragility). Just test the basics and pay the performance penalty (unlikely).

I prefer the first in an open business. But I can see some businesses hiding information like titles and descriptions.

(I love geek-i-tude)

Curt.