http://www.altuure.com/2008/11/12/object-model-vs-database-schema-design/
http://www.altuure.com/2008/10/22/why-orm-cause/

- You can fiddle with the used dialect if you find something wrong with the generated sql on a fundamental level.
- You can add sql snippets for calculating single attributes. Here combining analytic functions and hibernate works just great.
- You can define custom sql statements to be used for CRUD (http://hibernate.org/hib_docs/v3/reference/en/html/querysql-cud.html) I had to look up that myself
- you can obtain a jdbc connection and do whatever you like with it.

I am well aware of features like rollups, cubes and analytic functions. (e.g /2008/08/20/sql-tricks-3-intervall-join/). And they sure come in handy with reporting / datawarehouse style application.

But also in many cases the SQL skills on a java heavy team arn’t that great, so it is still beneficial to get 98% of the sql generated, working just fine. So you can use the extra time for tuning the queries, that really cause a problem.

Precision: I’m talking of databases with tables having millions of rows. Having optimized queries is a necessity.

To my knowledge :

1 – If you use ORM tools such as Hibernate and others, you have nearly no way to optimize the text of sql queries that are sent to the db engine. Just buy new hardware with more CPUs, more RAM and store your db in cache. I do not call this “a solution” but it may be ok for you, it depends on your cash.

2 – If you can customize the text of the sql queries, then you can unleash the power of your database engine (Oracle, SqlServer, MySql, …): they all have very powerful and very little-known capabilities such as analytical operators, cube operators and such.

You may also decide if, and when, argument binding with JDBC is a good thing or not (eg: on Oracle, binding is considered “good” on OLTP with 100′s or 1 000′s requests per second – but it can be considered “bad” in other environments such as web frontends, DW, … because the query optimizer can not make the full use of table and index meta-data such as histograms or skew of the values stored in tables and indexes – that makes the difference between 10 seconds query run time, or sub-second run time).

==> Please correct me if Hibernate permits easy customization and maintenance of the sql queries sent to the database.

For the ones who wish to know more on that subject, you may Google on “group by rollup”, “group by cube”, “group by grouping sets”, “rank”, “dense_rank”, “lag”, “lead”.

I work with Oracle but other database engines have the same sort of extensions – this is not Oracle-specific.

You may also read http://www.nocoug.org/download/2001-11/8i_analytical_features.ppt – this is based on a (very) old Oracle version but foundations still hold true.

IMPORTANT: as an exercise, at each slide, try to write the same query using only what is called “standard SQL” (i.e. without analytical operators): when dealing with LAG and LEAD operators for example, or grouped running totals, you may discover new horizons.