If you are just comparing based on your raw, input data, keep this in mind -- you need to buy at least 3, and probably closer to 5, times that much total disk size. It's true in all our implementations. Also take a look at www.tpc.org, specifically benchmarks TPC-H
Requires Free Membership to View
and TPC-R. They list combinations of hardware and DBMS that the vendors are
putting forward for benchmark purposes. In the disclosures, the "take up" from raw disk is
mentioned. Many of them go well beyond 5 times raw, but since none of us are implementing the
benchmarks and needing every ounce of performance possible at any price, 3-5 times is a safe
estimate.
The exact way to size is to know your architecture, all your tables and how many rows will be placed in each table for the next 3+ years. Rows per page can then be computed followed by the number of pages necessary. Of course, usually you need the system in place well before the time you know all this information, including the location of all indexes and summary tables. This precise approach is neither necessary nor feasible. Hence, the rules of thumb and the importance of getting into something scalable.
So, one important step in determining the DBMS for large-sized warehouses is to compare your profile to existing warehouses in production for the DBMS you are considering. When you do this, just make sure you compare apples-to-apples, understanding something about the take-up factors and the architectures of the systems that the vendors put forward to you for comparison.
For more information, check out SearchCRM's Best Web Links on Data Warehousing.
This was first published in May 2002
Join the conversationComment
Share
Comments
Results
Contribute to the conversation