Database Administrators Stack Exchange is a question and answer site for database professionals who wish to improve their database skills and learn from others in the community. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Would the following approach be reasonable?

The Goal: queries have to be relatively fast, but without the hassles incurred by writing to a de-normalized database.

The Context: A database for handling user profiles for some CMS. User profiles sometimes change, but are not updated too regularly; so for the most part the information for each user remains the same for some time.

The Database Design:

  • First, we put together a highly normalized database, call it "N". User profiles are stored in N.

  • Being highly normalized, N is secure and elegant to update, but slow to query. So we take all the data stored in N and we create a single de-normalized table. We call this table T.

  • T is used for the sole purpose of retrieving user information. It is only read from, not written to, because it's prone to anomalies and would be a mess to update.

With one exception:

  • To table T, we add a boolean column, "hasChanged", default value "FALSE".

The CMS's algorithm for handling queries:

Suppose we want to load user 47's profile.

  1. If hasChanged == FALSE for a row with ID == 47, we can safely retrieve the user profile from this row. (Because we know nothing has changed.)

  2. If hasChanged == TRUE, then row 47 in T cannot be trusted (something has changed), so the query is passed on to the reliable but slow N.

  3. When a user updates his profile, N is updated. When N is updated, hasChanged is set to TRUE on all rows in T containing information that could be affected by this change.

  4. T is completely rewritten by the system whenever it becomes too unreliable (this is determined by how frequently queries are passed on to N).

share|improve this question
up vote 3 down vote accepted

Use the full normalized database, scan which queries need more performance than this model can give them and create materialized views on them. If the data is very volatile, having them fast refreshing introduces some overhead but you know the value and validity of your data, that is protected by the database. Oracle handles this very nice.

You queries remain using the normalized tables and Query Rewrite will take care of the usage of the Materialized Views, when appropriate. When the Materialized Views are not fresh, the queries fall back to the original tables.

For docu see Materialized View Concepts and Architecture

share|improve this answer
Thanks, I'll read about materialized views. I'm still only familiar with the basic MAMP/LAMP server installation, so I wouldn't understand too many technical nuances. – iDontKnowBetter Aug 13 '12 at 21:09

"Slow to query" is essentially meaningless. But actual numbers from a representative prototype are meaningful.

If hasChanged == FALSE for a row with ID == 47, we can safely retrieve the user profile from this row. (Because we know nothing has changed.)

That's true only if there's a trigger on every table whose data is reflected in the denormalized table.

Application code can't reliably maintain this value, because your dbms probably has at least two interfaces--a command-line interface and a GUI interface--that let you change data without using the application code.

If you worked for me, I'd make you prove that your performance concerns are well-founded. Show me that you've

  • normalized most (if not all) tables to 5NF,
  • used both surrogate keys and natural keys wisely and judiciously,
  • studied the execution plan for relevant queries,
  • added indexes where they're needed,
  • considered the effect of "unusual" indexes where they might be useful,
  • considered the effect of caching,
  • tested performance on representative hardware under representative load (both in terms of concurrent users and in terms of row counts), and
  • considered other platform-specific features (like materialized views mentioned in other answers),
  • considered tuning the server,
  • considered upgrading the server's hardware.

That's off the top of my head, but that's most of the list. Some of these are impractical for some companies. (Representative hardware is sometimes prohibitively expensive.)

Surrogate integers always require a join. Often a human-readable code is a better choice. "Unusual" indexes include every kind of index the target dbms supports besides what you might get from a naive create index statement.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.