Tuning of SQL code is generally cheaper than changing the data model. Physical and configuration tuning involves a search for bottlenecks that often points to SQL code or data model issues. Building an appropriate data model and writing properly performing SQL code can give 100%+ performance improvement. Physical and configuration tuning often gives at most a 25% performance increase. Gavin Powell shows that the central theme of Oracle10gR2 Performance Tuning is four-fold: denormalize data models to fit applications; tune SQL code according to both the data model and the application in relation to scalability; create a well-proportioned physical architecture at the time of initial Oracle installation; and most important, mix skill sets to obtain the best results.