Re: Announcing the "Instant Data Warehouse" Product

NRaden wrote:
>
>
> We call this setup the "consolidated star" and it is classic Metaphor.
> Information Advantage seems to prefer this setup too and unless I'm
> mistaken, actually requires it (or at least a view that looks like it). My
> problem with it is that it slows down queries at aggregate levels, since
> we have to query a much large fact table, instead of a much smaller one.
> It also complicates the creation and maintenace of aggregates at load
> time, but it does vastly simplify the maintenance of metadata. Tradeoffs.
> Each situation needs to be evaluated by someone who understands the
> differences.
>

Neil,

IA does support the "consolidated star" model but we do not require it. In fact, most of our customers partition their fact tables by aggregate level. Our OLAP Engine’s SQL generation is influenced by the metadata model. If the metadata model shows that the fact tables are indeed partitioned by aggregate level, the SQL generated will zero in on the right table, thereby improving performance. This is seamless to the user. All the users see is a list of facts and calculations they can select from to use in their reports, regardless of how many fact tables physically exist.

Most of our customers have anywhere from 1 to 60 physical fact tables supporting data warehouses from 5 gigabytes to 1.2 terabytes. Like you said, this can increase the maintenance of the RDBMS because there are more tables, however many have implemented the partitioned tables to simplify load and recovery management - yet another trade off.    

I am intrigued by your comment on metadata maintenance. This is a trivial exercise with our product.
It amounts to a single entry in the metadata model to describe the location of the tables to the OLAP Engine. Once updated, all subsequent SQL generation will have knowledge of the new level. Metadata maintenance is only required when the aggregate levels of the fact table change. In our experience, this does not occur very often and the real work is updating the physical tables. Could you explain in more detail why you think this creates a large metadata maintenance problem.

Jim Frome
Information Advantage
jimf_at_infoadvan.com Received on Wed Feb 21 1996 - 00:00:00 CET