Archives of the TeradataForum
Message Posted: Mon, 05 Nov 2013 @ 00:57:44 GMT
Unfortunately, economists have a saying that the "bad money" chases out the "good money". This saying is in reference to cheap substitution occurring for the highest possible quality, as the cheap products cover 90-95% of the market's needs. This has occurred in the SCD field and so all SCD forms have collapsed around Type I & II, (plus a Type 0) and there are references to Type III and VI (6), a hybrid of the others and no references currently exist regarding the former Classes 1-6; at least I could find none. Unfortunely, in the process of a PC conversion, all my related documents were lost (sigh), and I could locate my previous references. If I find them, I will publish that. At any rate, no current approaches accurately represents the nature of what (and is) was Class 2, 4, 5 and 6.
Therefore, since no product/solution exists to address the Class 4, 5, and 6, the concept of the Classes has fallen silent. In this day of nationalized healthcare, "big data" and 5.2 Yottabyte data centers (see the NSA data center in Utah. A Yottabyte is 1 trillion terabytes), highly rigorous retention of Slowly Changing Dimensions will become increasingly critical. However, this critical need will remain unmet and data will become both less secure and of lower quality, as the unique identifiers of the massive data stores will become less reliable.
So my answer for this thread is that there are no current robust discussion of SCD Classes 4, 5 & 6.
Regarding our approach to Class 5, (we have not implemented a Class 6 solution, but could), I will indicate that at a high level, they consist of identifying the SCD's of each and every dimension, at least those of high value, and retaining the point in time (usually/currently a 24 hour period, but seconds to sub-seconds may become the critical time granularity) when the content of the dimension stores in scope (usually "columns") changes in a row, and storing that content and the date of the content, plus retaining the prior content. In our case, we retain not only the values of dimensional columns and their changes over time, plus the times of the those changes, we also retain the columns that comprise the DDL and those DDL changes. So if columns are added to or deleted from a table, that information is also retained. We can re-create the exact DDL of a table over time, plus the content of those columns. In actual practice, we can go back in time and re-create the tables and their content at any point in time.
Sorry for the long answer, but current SCD content on the Internet is focused on the SCD "types".
|Copyright 2016 - All Rights Reserved|
|Last Modified: 24 Jul 2020|