This is my first post, so I hope I am not infringing on any rules. We are using DB2 os/390 v7. We have several xref tables cross referencing corps, accts, custs, users, and cards. The Java developers have a need to join all tables for all needed data in one join so as to perform only one call and the levels of nested jons are becoming horrendous. What is the current opinion on xref tables and joins vs denormalization.
Good question! We are a OS/390 user and have a only a few examples of denormalization. The main point is how often underlying data will be updated. Best compromise is probably to have the both. All updates will be done to "normal" 3NL-normalized table and the triggers will update denormalized table used in queries.
We have 30 something big applications, now only 2 of them are using denormalized query table. Usually big joins are quite OK (if there are good indexes and same clustering key).