Re: Functions in the relational context

From: -CELKO- <>
Date: Sat, 8 Mar 2008 11:08:55 -0800 (PST)
Message-ID: <>

This is funny to me. I just wrote a book entitled "Joe Celko's Thinking in Sets: Auxiliary, Temporal, and Virtual Tables in SQL" which is about using tables in place of computational code for the working programmer.

My premise is that parallelism and fast, cheap storage (sold state devices replacing moving disk) make this approach faster and more portable than re-computing values over and over.

In real applications, the number of function values needed is (whatever "small" is this week). My favorite personal example is written up in "Data Mining on a Budget" ( /5343). I like it because
1) Computational approaches were too complex to work 2) I got paid in BBQ

When the auxiliary table does not have the value we want, then we can: 1) Throw an exception and let the application handle it. 2) Take time to compute it and insert the answer back into the table.
3) Interpolate the answer, knowing that we are trading accuracy for speed. This is often fine in statistical application.

I think it is a valid approach to keep SQL programming as declarative as possible. Received on Sat Mar 08 2008 - 20:08:55 CET

Original text of this message