Monday, October 1, 2007

Targetting the Database Architecture from Design

The following is an excerpt of a paper I am currently writing on Oracle performance tuning and related high availability strategies:

The Best Database Architectures are Designed

During my twenty years in information and database technology, I have encountered that the best database architecture designs were not directly implemented in a production environment, but instead they had been scaled in smaller servers, and systematically brought up to larger servers, in comparison to incremental prototypes and iterative software engineering methodologies. As such, I implemented my first multiple block size in 2G Solaris box, before I used in a 16GB server and later on in a three-level fault tolerant mainframe with four times the number of dual cores and RAM.

Significantly, the results showed that modeling architectures from design is a more effective way to achieve projected performance tuning in large scales, rather than having to deal with a potentially a very large number of events and factors in a production environment. Likewise, when database architectures have to be modified in an intermediate stage or already in production stage, quality assurance methodologies similar to those used in software engineering, as mentioned above, are applied to database technology performance optimization.

Similarly, I have stated in several related papers that about 80% of performance tuning is driving by the original target design, and this is more significant when it is already put into production. As such, no architectural component, application, I/O or SQL statement, and so on, can fully achieve optimal performance when the original design is neglected.

No comments: