How does CouchDB work in OLAP?

Question

I have a data table consisting of a large number (say, a billion) of vectors [ x1, x2, x3 ] , and I want to ask typical OLAP questions like "for all vectors with x1 in a given range, what is the average of v3?" Unlike true OLAP, questions are not ad-hoc: I only have a few predefined questions.

In the SQL database, you can say that if the columns are not indexed, the space requirements are O (n) as well as time; indexing gives you O (log n) for time, due to O (n log n) for space.

So, is CouchDB roughly equivalent in performance? Much better? Much worse?

Answer 1

CouchDB will usually be worse if you want to make special requests, and better if you have pre-baked requests.

This is just a technical point about CouchDB, not NOSQL and SQL.

CouchDB is a bit slower in ad-hoc requests, which I think require table scans. But views are gradually updated as you gradually add data, so it’s useful to support things like “Sum”, “Count” or something else that can be reduced with a map.