Redshift is a sorted relational database and as such it provides the performance necessary to handle Big Data when and only operated correctly, which is to say, when the sorting orders of the tables are correct for the queries being issued on those tables.

If you start talking as the quotes here do about "extremely complex SQL" it is going to be utterly impractical to operate Redshift correctly; it's impossibly complicated.

When operated incorrectly, Redshift has the same performance profile as good old Postgres, except you have the cluster rather than a single node.

The cluster is an expensive way to get 10x or 100x performance, but if you have Big Data, you need a LOT more performance than this - you need sorting, with the staggering efficiency gains it brings - but sorting only works when operated correctly.

The docs from AWS here are, in my view, utterly misleading and are dangerous to readers.